Introduction to Algebraic Geometry

This page intentionally left blank Introduction to Algebraic Geometry Algebraic geometry has a reputation for being d...

Author: Brendan Hassett

30 downloads 1572 Views 1MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form

DOWNLOAD PDF

This page intentionally left blank

Introduction to Algebraic Geometry Algebraic geometry has a reputation for being difficult and inaccessible, even among mathematicians! This must be overcome. The subject is central to pure mathematics, and applications in fields like physics, computer science, statistics, engineering, and computational biology are increasingly important. This book is based on courses given at Rice University and the Chinese University of Hong Kong, introducing algebraic geometry to a diverse audience consisting of advanced undergraduate and beginning graduate students in mathematics, as well as researchers in related fields. For readers with a grasp of linear algebra and elementary abstract algebra, the book covers the fundamental ideas and techniques of the subject and places these in a wider mathematical context. However, a full understanding of algebraic geometry requires a good knowledge of guiding classical examples, and this book offers numerous exercises fleshing out the theory. It introduces Gröbner bases early on and offers algorithms for almost every technique described. Both students of mathematics and researchers in related areas benefit from the emphasis on computational methods and concrete examples. Brendan Hassett is Professor of Mathematics at Rice University, Houston

Introduction to Algebraic Geometry Brendan Hassett Department of Mathematics, Rice University, Houston

CAMBRIDGE UNIVERSITY PRESS

Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, São Paulo Cambridge University Press The Edinburgh Building, Cambridge CB2 8RU, UK Published in the United States of America by Cambridge University Press, New York www.cambridge.org Information on this title: www.cambridge.org/9780521870948 © Brendan Hassett 2007 This publication is in copyright. Subject to statutory exception and to the provision of relevant collective licensing agreements, no reproduction of any part may take place without the written permission of Cambridge University Press. First published in print format 2007 eBook (EBL) ISBN-13 978-0-511-28529-5 ISBN-10 0-511-28529-9 eBook (EBL) hardback ISBN-13 978-0-521-87094-8 hardback ISBN-10 0-521-87094-1 paperback ISBN-13 978-0-521-69141-3 paperback ISBN-10 0-521-69141-9 Cambridge University Press has no responsibility for the persistence or accuracy of urls for external or third-party internet websites referred to in this publication, and does not guarantee that any content on such websites is, or will remain, accurate or appropriate.

To Eileen and William

Contents

Preface

xi

1

Guiding problems 1.1 Implicitization 1.2 Ideal membership 1.3 Interpolation 1.4 Exercises

1 1 4 5 8

2

Division algorithm and Gröbner bases 2.1 Monomial orders 2.2 Gröbner bases and the division algorithm 2.3 Normal forms 2.4 Existence and chain conditions 2.5 Buchberger’s Criterion 2.6 Syzygies 2.7 Exercises

11 11 13 16 19 22 26 29

3

Aff ine varieties 3.1 Ideals and varieties 3.2 Closed sets and the Zariski topology 3.3 Coordinate rings and morphisms 3.4 Rational maps 3.5 Resolving rational maps 3.6 Rational and unirational varieties 3.7 Exercises

33 33 38 39 43 46 50 53

4

Elimination 4.1 Projections and graphs 4.2 Images of rational maps 4.3 Secant varieties, joins, and scrolls 4.4 Exercises

57 57 61 65 68 vii

viii

CONTENTS

5

Resultants 5.1 Common roots of univariate polynomials 5.2 The resultant as a function of the roots 5.3 Resultants and elimination theory 5.4 Remarks on higher-dimensional resultants 5.5 Exercises

73 73 80 82 84 87

6

Irreducible varieties 6.1 Existence of the decomposition 6.2 Irreducibility and domains 6.3 Dominant morphisms 6.4 Algorithms for intersections of ideals 6.5 Domains and field extensions 6.6 Exercises

89 90 91 92 94 96 98

7

Nullstellensatz 7.1 Statement of the Nullstellensatz 7.2 Classification of maximal ideals 7.3 Transcendence bases 7.4 Integral elements 7.5 Proof of Nullstellensatz I 7.6 Applications 7.7 Dimension 7.8 Exercises

101 102 103 104 106 108 109 111 112

8

Primary decomposition 8.1 Irreducible ideals 8.2 Quotient ideals 8.3 Primary ideals 8.4 Uniqueness of primary decomposition 8.5 An application to rational maps 8.6 Exercises

116 116 118 119 122 128 131

9

Projective geometry 9.1 Introduction to projective space 9.2 Homogenization and dehomogenization 9.3 Projective varieties 9.4 Equations for projective varieties 9.5 Projective Nullstellensatz 9.6 Morphisms of projective varieties 9.7 Products 9.8 Abstract varieties 9.9 Exercises

134 134 137 140 141 144 145 154 156 162

CONTENTS

ix

10

Projective elimination theory 10.1 Homogeneous equations revisited 10.2 Projective elimination ideals 10.3 Computing the projective elimination ideal 10.4 Images of projective varieties are closed 10.5 Further elimination results 10.6 Exercises

169 170 171 174 175 176 177

11

Parametrizing linear subspaces 11.1 Dual projective spaces 11.2 Tangent spaces and dual varieties 11.3 Grassmannians: Abstract approach 11.4 Exterior algebra 11.5 Grassmannians as projective varieties 11.6 Equations for the Grassmannian 11.7 Exercises

181 181 182 187 191 197 199 202

12

Hilbert polynomials and the Bezout Theorem 12.1 Hilbert functions defined 12.2 Hilbert polynomials and algorithms 12.3 Intersection multiplicities 12.4 Bezout Theorem 12.5 Interpolation problems revisited 12.6 Classification of projective varieties 12.7 Exercises

207 207 211 215 219 225 229 231

Appendix A Notions from abstract algebra A.1 Rings and homomorphisms A.2 Constructing new rings from old A.3 Modules A.4 Prime and maximal ideals A.5 Factorization of polynomials A.6 Field extensions A.7 Exercises

235 235 236 238 239 240 242 244

Bibliography Index

246 249

Preface

This book is an introduction to algebraic geometry, based on courses given at Rice University and the Institute of Mathematical Sciences of the Chinese University of Hong Kong from 2001 to 2006. The audience for these lectures was quite diverse, ranging from second-year undergraduate students to senior professors in fields like geometric modeling or differential geometry. Thus the algebraic prerequisites are kept to a minimum: a good working knowledge of linear algebra is crucial, along with some familiarity with basic concepts from abstract algebra. A semester of formal training in abstract algebra is more than enough, provided it touches on rings, ideals, and factorization. In practice, motivated students managed to learn the necessary algebra as they went along. There are two overlapping and intertwining paths to understanding algebraic geometry. The first leads through sheaf theory, cohomology, derived functors and categories, and abstract commutative algebra – and these are just the prerequisites! We will not take this path. Rather, we will focus on specific examples and limit the formalism to what we need for these examples. Indeed, we will emphasize the strand of the formalism most useful for computations: We introduce Gröbner bases early on and develop algorithms for almost every technique we describe. The development of algebraic geometry since the mid 1990s vindicates this approach. The term ‘Groebner’ occurs in 1053 Math Reviews from 1995 to 2004, with most of these occurring in the last five years. The development of computers fast enough to do significant symbolic computations has had a profound influence on research in the field. A word about what this book will not do: We develop computational techniques as a means to the end of learning algebraic geometry. However, we will not dwell on the technical questions of computability that might interest a computer scientist. We will also not spend time introducing the syntax of any particular computer algebra system. However, it is necessary that the reader be willing to carry out involved computations using elementary algebra, preferably with the help of a computer algebra system such as Maple, Macaulay II, or Singular. Our broader goal is to display the core techniques of algebraic geometry in their natural habitat. These are developed systematically, with the necessary commutative algebra integrated with the geometry. Classical topics like resultants and elimination xi

xii

PREFACE

theory, are discussed in parallel with affine varieties, morphisms, and rational maps. Important examples of projective varieties (Grassmannians, Veronese varieties, Segre varieties) are emphasized, along with the matrix and exterior algebra needed to write down their defining equations. It must be said that this book is not a comprehensive introduction to all of algebraic geometry. Shafarevich’s book [37, 38] comes closest to this ideal; it addresses many important issues we leave untouched. Most other standard texts develop the material from a specific point of view, e.g., sheaf cohomology and schemes (Hartshorne [19]), classical geometry (Harris [17]), complex algebraic differential geometry (Griffiths and Harris [14]), or algebraic curves (Fulton [11]).

Acknowledgments I am grateful to Ron Goldman, Donghoon Hyeon, Frank Jones, Sándor Kovács, Manuel Ladra, Dajiang Liu, Miles Reid, Burt Totaro, Yuri Tschinkel, and Fei Xu for helpful suggestions and corrections. I am indebted to Bradley Duesler for his comments on drafts of the text. My research has been supported by the Alfred P. Sloan Foundation and the National Science Foundation (DMS 0554491, 0134259, and 0196187). The treatment of topics in this book owes a great deal to my teachers and the fine textbooks they have written: Serge Lang [27], Donal O’Shea [8], Joe Harris [17], David Eisenbud [9], and William Fulton [11]. My first exposure to algebraic geometry was through drafts of [8]; it has had a profound influence on how I teach the subject.

1 Guiding problems

Let k denote a field and k[x1 , x2 , . . . , xn ] the polynomials in x1 , x2 , . . . , xn with coefficients in k. We often refer to k as the base field. A nonzero polynomial f =

α1 ,...,αn

cα1 ...αn x1α1 . . . xnαn ,

cα1 ...αn ∈ k,

has degree d if cα1 ...αn = 0 when α1 + · · · + αn > d and cα1 ...αn = 0 for some index with α1 + · · · + αn = d. It is homogeneous if cα1 ...αn = 0 whenever α1 + · · · + αn < d. We will sometimes use multiindex notation f =

α

cα x α

where α = (α1 , . . . , αn ), cα = cα1 ...αn , x α = x1α1 . . . xnαn , and |α| = α1 + · · · + αn ,

1.1

Implicitization Definition 1.1

Affine space of dimension n over k is defined An (k) = {(a1 , a2 , . . . , an ) : ai ∈ k}.

For k = R this is just the ubiquitous Rn . Why don’t we use the notation k n for affine space? We write An (k) when we want to emphasize the geometric nature of k n rather than its algebraic properties (e.g., as a vector space). Indeed, when our discussion does not involve the base field in an essential way we drop it from the notation, writing An . We shall study maps between affine spaces, but not just any maps are allowed in algebraic geometry. We consider only maps given by polynomials: Definition 1.2

A morphism of affine spaces φ : An (k) → Am (k) 1

2

GUIDING PROBLEMS

is a map given by a polynomial rule (x1 , x2 , . . . , xn ) → (φ1 (x1 , . . . , xn ), . . . , φm (x1 , . . . , xn )), with the φi ∈ k[x1 , . . . , xn ]. This makes a tacit reference to the base field k, in that the polynomials φi have coefficients in k. If we want to make this explicit, we say that the morphism is defined over k.

Remark 1.3

An affine-linear transformation is a morphism: given an m × n matrix A = (ai j ) and an m × 1 matrix b = (bi ) with entries in k, we define

Example 1.4

φ A,b : An (k) → Am (k) ⎛ ⎞ ⎛ ⎞ x1 a11 x1 + · · · + a1n xn + b1 ⎜ .. ⎟ ⎟ ⎜ .. ⎝ . ⎠ → ⎝ ⎠. . xn Example 1.5

am1 x1 + · · · + amn xn + bm

Consider A1 (R) → A2 (R)

given by the rule t → (t, t 2 ). If y1 and y2 are the corresponding coordinates on R2 then the image is the parabola {(y1 , y2 ) : y2 = y12 }. More generally, consider the morphism φ : A1 (k) → Am (k) t → (t, t 2 , t 3 , . . . , t m ). Can we visualize the image of φ in Am (k)? Just as for the parabola, we write down polynomial equations for this locus. Fix coordinates y1 , . . . , ym on Am (k) so that φ is given by yi → t i . We find the equations yi y j = yi+ j yi y j = yk yl

1≤i < j ≤m i + j = k +l

corresponding to the relations t i t j = t i+ j and t i t j = t k t l respectively. The polynomial equations describing the image of our morphism are an implicit description of this locus. Here the sense of ‘implicit’ is the same as the ‘implicit function theorem’ from calculus. We can consider the general question: Problem 1.6 (Implicitization)

the image of a morphism.

Write down the polynomial equations satisfied by

1.1 IMPLICITIZATION

1.1.1 A special case: linear transformations

3

Elementary row operations from linear algebra solve Problem 1.6 in the case where φ is a linear transformation. Suppose φ is given by the rule A2 (Q) → A3 (Q) (x1 , x2 ) → (x1 + x2 , x1 − x2 , x1 + 2x2 ) and assign coordinates y1 , y2 , y3 to affine three-space. From this, we extract the system y1 = x1 + x2 y2 = x1 − x2 y3 = x1 + 2x2 , or equivalently, x1 +x2 −y1 =0 x1 −x2 −y2 =0 x1 +2x2 −y3 = 0, which in turn are equivalent to x1 +x2 −y1 =0 =0 −2x2 +y1 −y2 x2 +y1 −y3 = 0, and x1 +x2 −y1 =0 −2x2 +y1 −y2 =0 + 32 y1 − 12 y2 −y3 = 0. Thus the image of our morphism is given by 3y1 − y2 − 2y3 = 0. Our key tool for solving Problem 1.6 in general – Buchberger’s Algorithm – will contain elementary row operations as a special case. Moral 1: To solve Problem 1.6, choosing an order on the variables is very useful.

1.1.2 A converse to implicitization?

The implicitization problem seeks equations for the image of a morphism φ : An (k) → Am (k). We will eventually show that this admits an algorithmic solution, at least when the base field is algebraically closed. However, there is a natural converse to this question which is much deeper.

4

GUIDING PROBLEMS

Definition 1.7

A hypersurface of degree d is the locus

V ( f ) := {(a1 , . . . , am ) ∈ Am (k) : f (a1 , . . . , am ) = 0} ⊂ Am (k), where f is a polynomial of degree d. A regular parametrization of a hypersurface V ( f ) ⊂ Am (C) is a morphism φ : An (C) → Am (C) such that 1. 2.

1. 2.

the image of φ is contained in the hypersurface, i.e., f ◦ φ = 0; the image of φ is not contained in any other hypersurface, i.e., for any h ∈ C[y1 , . . . , ym ] with h ◦ φ = 0 we have f |h. Problem 1.8

Which hypersurfaces admit regular parametrizations?

Example 1.9

Here are some cases where parametrizations exist:

hypersurfaces of degree one (see Exercise 1.5); the curve V ( f ) ⊂ A2 , f = y12 − y23 , has parametrization (cf. Exercise 1.8) φ : A1 (C) → A2 (C) t → (t 3 , t 2 )

3.

if f = y02 + y12 − y22 then V ( f ) has a parametrization φ(s, t) = (2st, s 2 − t 2 , s 2 + t 2 );

4.

if f = y03 + y13 + y23 + y33 then V ( f ) has parametrization

y0 = (u 2 + u 1 )u 23 + u 22 + 2u 21 u 3 − u 32 + u 1 u 22 − 2u 21 u 2 − u 31

y1 = u 33 − (u 2 + u 1 )u 23 + u 22 + 2u 21 u 3 + u 1 u 22 − 2u 21 u 2 + u 31

y2 = −u 33 + (u 2 + u 1 )u 23 − u 22 + 2u 21 u 3 + 2u 1 u 22 − u 21 u 2 + 2u 31

y3 = (u 2 − 2u 1 )u 23 + u 21 − u 22 u 3 + u 32 − u 1 u 22 + 2u 21 u 2 − 2u 31 . The form here is due to Noam Elkies. We will come back to these questions when we discuss unirationality and rational maps in Chapter 3.

1.2

Ideal membership Our second guiding problem is algebraic in nature. Given f 1 , . . . , fr ∈ k[x1 , . . . , xn ], determine whether g ∈ k[x1 , . . . , xn ] belongs to the ideal f 1 , . . . , fr .

Problem 1.10 (Ideal Membership Problem)

1.3 INTERPOLATION

Example 1.11

5

Consider the ideal I = y2 − y12 , y3 − y1 y2 ⊂ k[y1 , y2 , y3 ]

and the polynomial g = y1 y3 − y22 (cf. Example 1.5 and the following discussion). Then g ∈ I because

y1 y3 − y22 = y1 (y3 − y1 y2 ) + y2 y12 − y2 . Again, whenever the f i and g are all linear, elementary row reductions give a solution to Problem 1.10. However, there is one further case where we already know how to solve the problem. The Euclidean Algorithm yields a procedure to decide whether a polynomial g ∈ k[t] is contained in a given ideal I ⊂ k[t]. By Theorem A.9, each ideal I ⊂ k[t] can be expressed I = f for some f ∈ k[t]. Therefore g ∈ I if and only if f divides g. Example 1.12

Check whether t 5 + t 3 + 1 ∈ t 3 + 1 : t 2 +1 t + 1 t 5 +t 3 +1 t5 +t 2 3

+t 3 −t 2 +1 +t 3 +1 −t 2 thus q = t 2 + 1 and r = −t 2 . We conclude t 5 + t 3 + 1 ∈ t 3 + 1 : Moral 2: In solving Problem 1.10, keeping track of degrees of polynomials is crucial.

1.3

Interpolation Let Pn,d ⊂ k[x1 , . . . , xn ] denote the vector subspace of polynomials of degree ≤ d. The monomials x α = x1α1 . . . xnαn ,

α1 + · · · + α n ≤ d

form a basis for Pn,d , so we have (see Exercise 1.4)

n+d dim Pn,d = . n Problem 1.13 (Simple Interpolation Problem)

Given distinct points

p1 , . . . , p N ∈ An (k) what is the dimension of the vector space Id ( p1 , . . . , p N ) of polynomials of degree ≤ d vanishing at each of the points?

6

GUIDING PROBLEMS

Here is some common terminology used in these questions: Given S ⊂ An (k), the number of conditions imposed by S on polynomials of degree ≤ d is defined

Definition 1.14

Cd (S) := dim Pn,d − dim Id (S). S is said to impose independent conditions on Pn,d if Cd (S) = |S|. It fails to impose independent conditions otherwise. Another formulation of the Simple Interpolation Problem is: When do N points in An (k) fail to impose independent conditions on polynomials of degree ≤ d?

In analyzing examples, it is useful to keep in mind that affine linear transformations do not affect the number conditions imposed on Pn,d : Let S ⊂ An (k) and consider an invertible affine-linear transformation φ : A (k) → An (k). Then Cd (S) = Cd (φ(S)) for each d.

Proposition 1.15 n

By Exercise 1.11, φ induces an invertible linear transformation φ ∗ : Pn,d → Pn,d with φ ∗ ( f (x1 , . . . , xn )) = ( f ◦ φ)(x1 , . . . , xn ). Thus (φ ∗ f )( p) = 0 for each p ∈ S if and only if f (q) = 0 for each q ∈ φ(S). In particular, φ ∗ (Id (φ(S))) = Id (S) so these spaces have the same dimension.

Proof

1.3.1 Some examples

Let S = { p1 , p2 , p3 } ⊂ An (k) be collinear with n > 1 or S = { p1 , p2 , p3 , p4 } ⊂ An (k) coplanar with n > 2. Then S fails to impose independent conditions on polynomials of degree ≤ 1. Let S = { p1 , p2 , p3 , p4 , p5 , p6 } ⊂ A2 (R) lie on the unit circle x12 + x22 = 1. Then S fails to impose independent conditions on polynomials of degree ≤ 2; indeed, C2 (S) = 5 < 6. When does a set of four points { p1 , p2 , p3 , p4 } ⊂ A2 (k) fail to impose independent conditions on quadrics (d = 2)? Assume that three of the points are non-collinear, e.g., p1 , p2 , p3 . After translating suitably we may assume p1 = (0, 0), and after a further linear change of coordinates we may assume p2 = (1, 0) and p3 = (0, 1). (Proposition 1.15 allows us to change coordinates without affecting the number of conditions imposed.) If p4 = (a1 , a2 ) then the conditions on c00 + c10 x1 + c01 x2 + c20 x12 + c11 x1 x2 + c02 x22 ∈ P2,2

1.3 INTERPOLATION

7

take the form c00 c00 + c10 + c20 c00 + c01 + c02 c00 + c10 a1 + c01 a2 + c20 a12 + c11 a1 a2 + c02 a22

= = = =

0 0 0 0.

( p1 ) ( p2 ) ( p3 ) ( p4 )

If these are not independent, the matrix ⎛

1 ⎜0 ⎜ ⎝0 0

0 1 0 0

0 0 0 1 1 0 0 a12 − a1

0 0 0 a1 a2

⎞ 0 0 ⎟ ⎟ 1 ⎠ a22 − a2

has rank 3. This can only happen if a12 − a1 = a1 a2 = a22 − a2 = 0, which means p4 ∈ {(0, 0), (1, 0), (0, 1)} = { p1 , p2 , p3 }, a contradiction. Thus we have shown: Four distinct points in the plane fail to impose independent conditions on quadrics only if they are all collinear.

Proposition 1.16

Here are some sample results: Any N points in the affine line A1 (k) impose independent conditions on P1,d for d ≥ N − 1.

Assume k is infinite. For each N ≤ n+d , there exist N points in An (k) imposing d independent conditions on Pn,d .

Proposition 1.17

Proof

For the first statement, suppose that f ∈ k[x1 ] is a polynomial vanishing

at p1 , . . . , p N ∈ A1 (k). The Euclidean Algorithm implies that f is divisible by x − p j for each j = 1, . . . , N . Consequently, it is also divisible by the product (x1 − p1 ) . . . (x1 − p N ) (see Exercise A.13). Moreover, if f = 0 we have a unique expression f = q(x1 − p1 ) . . . (x1 − p N ),

q ∈ P1,d−N .

The polynomials of this form (along with 0) form a vector space of dimension d − N + 1, so Cd ( p1 , . . . , p N ) = min(N , d + 1).

8

GUIDING PROBLEMS

The second statement is established by producing a sequence of points p1 , . . . , p(n+d ) such that d

Id ( p1 , . . . , p j ) Id ( p1 , . . . , p j+1 )

for each j < n+d . The argument proceeds by induction. Given p1 , . . . , p j , linear d algebra gives a nonzero f ∈ Pn,d with f ( p1 ) = . . . = f ( p j ) = 0. It suffices to find some p j+1 ∈ An (k) such that f ( p j+1 ) = 0, which follows from the fact (Exercise 1.9) that every nonzero polynomial over an infinite field takes a nonzero value somewhere in An (k).

1.4

Exercises

1.1

Consider the linear morphism φ : A3 (R) → A4 (R) (t1 , t2 , t3 ) → (3t1 + t3 , t2 + 4t3 , t1 + t2 + t3 , t1 − t2 − t3 ).

1.2

Describe image(φ) as the locus where a linear polynomial vanishes. Decide whether g = t 3 + t 2 − 2 is contained in the ideal t 3 − 1, t 5 − 1 ⊂ Q[t]. If so, produce h 1 , h 2 ∈ Q[t] such that g = h 1 (t 3 − 1) + h 2 (t 5 − 1).

1.3

1.4

1.5

Consider the ideal I = y2 − y12 , y3 − y1 y2 , . . . , ym − y1 ym−1 ⊂ k[y1 , . . . , ym ]. Show this contains all the polynomials yi+ j − yi y j and yi y j − yk yl where i + j = k + l (cf. Example 1.5.) Show that the dimension of the vector space of polynomials of degree ≤ d in n variables is equal to the binomial coefficient

n+d (n + d)! = . d d! n! Compute the dimension of the vector space of homogeneous polyonomials of degreee d in n + 1 variables. Given f = c1 x1 + c2 x2 + · · · + cn xn + c0 with ci = 0 for some i > 0, exhibit a morphism φ : An−1 → An such that image(φ) = V ( f ) and φ is one-to-one.

1.4 EXERCISES

1.6

9

Let A = (ai j ) be an m × n matrix with entries in k and b = (b1 , . . . , bn ) ∈ k n . For each i = 1, . . . , m, set f i = ai1 x1 + · · · + ain xn ∈ k[x1 , . . . , xn ]

1.7

and g = b1 x1 + · · · + bn xn . Show that g ∈ f 1 , . . . , f m if and only if b is contained in the span of the rows of A. Consider the morphism j : A3 (k) → A6 (k) (u, v, w) → (u 2 , uv, v 2 , vw, w 2 , uw). Let a11 , a12 , a22 , a23 , a33 , and a13 be the corresponding coordinates on A6 (k) and ⎛ ⎞ a11 a12 a13 A = ⎝ a12 a22 a23 ⎠ a13

a23

a33

the symmetric matrix with these entries. (a) Show that the image of j satisfies the equations given by the two-by-two minors of A. (b) Compute the dimension of the vector space V in R = k[a11 , a12 , a22 , a23 , a33 , a13 ]

1.8 1.9 1.10

spanned by these two-by-two minors. (c) Show that every homogeneous polynomial of degree 2 in R vanishing on the image of j is contained in V . Hint: Degree-2 polynomials in R yield degree-4 polynomials in k[u, v, w]. Count dimensions! Show that the parametrization given for the curve V ( f ) ⊂ A2 (C), f = x12 − x23 satisfies the required properties. Let k be an infinite field. Suppose that f ∈ k[x1 , . . . , xn ] is nonzero. Show there exists a = (a1 , . . . , an ) ∈ An (k) with f (a1 , . . . , an ) = 0. Let S ⊂ An (k) be a finite nonempty subset and let k[S] denote the ring of k-valued functions on S. Show that the linear transformation Pn,d → k[S] f → f |S

1.11

is surjective if and only if S imposes independent conditions on polynomials of degree d. Let φ : An (k) → Am (k) be an affine linear transformation given by the matrix formula φ(x) = Ax + b (see Example 1.4). Consider the map induced by composition of polynomials φ ∗ : k[y1 , . . . , ym ] → k[x1 , . . . , xn ] P(y) → P(Ax + b).

10

GUIDING PROBLEMS

1.12 1.13

Show that (a) φ ∗ takes polynomials of degree ≤ d to polynomials of degree ≤ d; (b) φ is a k-algebra homomorphism; (c) if the matrix A is invertible then so is φ ∗ . Moreover, in case (c) the induced linear transformation φ ∗ : Pn,d → Pn,d is also invertible. Consider five distinct points in A2 (R) that fail to impose independent conditions on P2,3 . Show that these points are collinear, preferably by concrete linear algebra. Show that d + 1 distinct points p1 , . . . , pd+1 ∈ An (Q)

1.14

always impose independent conditions on polynomials in Pn,d . Let 1 , 2 , 3 be arbitrary lines in A3 (Q). (By definition, a line ⊂ A3 is the locus where two consistent independent linear equations are simultaneously satisfied, e.g., x1 + x2 + x3 − 1 = x1 − x2 + 2x3 − 4 = 0.) Show there exists a nonzero polynomial f ∈ P3,2 such that f vanishes on 1 , 2 , and 3 . Optional Challenge: Assume that 1 , 2 , and 3 are pairwise skew. Show that f is unique up to scalar.

2 Division algorithm and Gröbner bases

In this chapter we solve the Ideal Membership Problem for polynomial ideals. The key tool is Gröbner bases: producing a Gröbner basis for a polynomial ideal is analogous to putting a system of linear equations in row echelon form. Once we have a Gröbner basis, a multivariate division algorithm can be applied to decide whether a given polynomial sits in our ideal. We also discuss normal forms for polynomials modulo ideals. The existence of a Gröbner base can be deduced from nonconstructive arguments, but actually finding one can be challenging computationally. Buchberger’s Algorithm gives a general solution. The proof that it works requires a systematic understanding of the ‘cancellations’ among polynomials, which are usually called syzygies.

2.1

Monomial orders As we have seen, in order to do calculations we need a system for ordering the terms of a polynomial. For polynomials in one variable, the natural order is by degree, i.e., β

x1α > x1

if α > β.

However, for linear polynomials in many variables, we have seen that the order is essentially arbitrary. We first fix terminology. Given a polynomial cα1 ...αn x1α1 . . . xnαn each cα1 ...αn x1α1 . . . xnαn is a term. A polynomial of the form x α = x1α1 . . . xnαn is called a monomial. A monomial order > on k[x1 , . . . , xn ] is a total order on monomials satisfying the following:

Definition 2.1

11

¨ DIVISION ALGORITHM AND GR OBNER BASES

12

1. 2.

Multiplicative property If x α > x β then x α x γ > x β x γ (for any α, β, γ ). Well ordering An arbitrary set of monomials {x α }α∈A has a least element. The stipulation that > be a total order means that any monomials x α and x β are comparable in the order, i.e., either x α > x β , x α < x β , or x α = x β . The well-ordering condition is equivalent to the requirement that any decreasing sequence of monomials

Remark 2.2

x α(1) > x α(2) > x α(3) > . . . eventually terminates. We give some basic examples of monomial orders: This is basically the order on words in a dictionary. We have x α >lex x β if the first nonzero entry of (α1 − β1 , α2 − β2 , . . . , αn − βn ) is positive. For example, we have

Example 2.3 (Pure lexicographic order)

x1 >lex x23 >lex x2 x3 >lex x3100 . We prove this is a monomial order. Any two monomials are comparable: given x α and x β , either some α j − β j = 0 (in which case x α >lex x β or x α lex x β x γ if and only if x α >lex x β . Finally, given any set of monomials {x α }α∈A , we extract the smallest element. Consider the descending sequence of subsets A = A0 ⊃ A1 ⊃ A2 ⊃ . . . ⊃ An defined recursively by A j = {α ∈ A j−1 : α j is minimal}. Each element of A j is smaller (with respect to >lex ) than all the elements of A \ A j . On the other hand, An has a unique element, which is therefore the minimal element in A.

¨ 2.2 GR OBNER BASES AND THE DIVISION ALGORITHM

13

Fix a monomial order on k[x1 , . . . , xn ] and consider a nonzero

Definition 2.4

polynomial f =

α

cα x α .

The leading monomial of f (denoted LM( f )) is the largest monomial x α such that cα = 0. The leading term of f (denoted LT( f )) is the corresponding term cα x α . For instance, in lexicographic order the polynomial f = 5x1 x2 + 7x25 + 19x317 has leading monomial LM( f ) = x1 x2 and leading term LT( f ) = 5x1 x2 . One nonintuitive aspect of lexicographic order is that the degree of the terms is not paramount: the smallest degree term could be the leading one. We can remedy this easily: Example 2.5 (Graded lexicographic order) α

β

α

β

x α >grlex x β if deg(x α ) > deg(x β ) or

deg(x ) = deg(x ) and x >lex x . x α >grelex x β if deg(x α ) > deg(x β ) or deg(x α ) = deg(x β ) and the last nonzero α j − β j < 0. (Yes, this inequality goes the right way!) Note that x1 x2 x4 >grlex x1 x32 but x1 x2 x4 on k[x1 , . . . , xn ] and nonzero polynomials f 1 , . . . , fr ∈ k[x1 , . . . , xn ]. Given g ∈ k[x1 , . . . , xn ], we want to determine whether g ∈ f 1 , . . . , fr :

Algorithm 2.7 (Division procedure)

Put g0 = g. If there exists no f j with LM( f j )|L M(g0 ) then we STOP. Otherwise, pick such an f j0 and cancel leading terms by putting

Step 0

g1 = g0 − f j0 LT(g0 )/LT( f j0 ). ... Given gi , if there exists no f j with LM( f j )|LM(gi ) then we STOP. Otherwise, pick such an f ji and cancel leading terms by putting

Step i

gi+1 = gi − f ji LT(gi )/LT( f ji ). As we are cancelling leading terms at each stage, we have LM(g) = LM(g0 ) > LM(g1 ) > . . . > LM(gi ) > LM(gi+1 ) > . . . .

(2.1)


14

By the well-ordering property of the monomial order, such a chain of decreasing monomials must eventually terminate. If this procedure does not stop, then we must have g N = 0 for some N . Back-substituting using Equation 2.1, we obtain g=

N −1

f ji LT(gi )/LT( f ji ) =

i=0

r j=1

LT(gi )/LT( f ji )

ji = j

fj =

r

h j fj,

j=1

where the last sum is obtained by regrouping terms. Unfortunately, this procedure often stops prematurely. Even when g ∈ f 1 , . . . , fr , it may happen that LM(g) is not divisible by any LM( f j ). Example 2.8

1. 2.

Let f 1 = x + 1, f 2 = x and g = 1. We certainly have g ∈ f 1 , f 2 but LM(g) is not divisible by LM( f 1 ) or LM( f 2 ), so the procedure stops at the initial step. If f 1 = x + 2y + 1, f 2 = x − y − 5, and g = y + 2 then we have the same problem. Linear algebra presents a solution: Our system of equations corresponds to the augmented matrix

1 2 1 −1

1 . −5

Put this matrix in ‘row echelon form’ by subtracting the first row from the second

1 2 0 −3

1 , −6

which corresponds to the new set of generators f 1 , f˜2 = −3y − 6. Our division algorithm works fine for these new generators. To understand better why this breakdown occurs, we make the following definitions: A monomial ideal J ⊂ k[x1 , . . . , xn ] is an ideal generated by a collection of monomials {x α }α∈A .

Definition 2.9

The main example is the ideal of leading terms of an arbitrary ideal I ⊂ k[x1 , . . . , xn ]. Fix a monomial order > and let I ⊂ k[x1 , . . . , xn ] be an ideal. The ideal of leading terms is defined

Definition 2.10

LT(I ) := LT(g) : g ∈ I . By convention, LT(0) = 0.

¨ 2.2 GR OBNER BASES AND THE DIVISION ALGORITHM

15

Fix a monomial order > and let I ⊂ k[x1 , . . . , xn ] be an ideal. A Gröbner basis for I is a collection of nonzero polynomials

Definition 2.11

{ f 1 , . . . , fr } ⊂ I such that LT( f 1 ), . . . , LT( fr ) generate LT(I ). Nothing in the definition says that a Gröbner basis actually generates I ! We prove this a posteriori. Remark 2.12

Every generator for a principal ideal is a Gröbner basis.

Let I ⊂ k[x1 , . . . , xn ] be an ideal and f 1 , . . . , fr a Gröbner basis for I . The Division Algorithm terminates in a finite number of steps, with either gi = 0 or LT(gi ) not divisible by any of the leading terms LT( f j ).

Proposition 2.13

1.

In the first case, it returns a representation g = h 1 f 1 + · · · + h r fr

2.

h j ∈ k[x1 , . . . , xn ],

and g ∈ I . In the second case, we obtain an expression g = h 1 f 1 + · · · + h r fr + gi

LT(gi ) ∈ LT( f 1 ), . . . , LT( fr ),

hence g ∈ I . The proposition immediately implies the following corollary. Fix a monomial order >. Let I ⊂ k[x1 , . . . , xn ] be an ideal and f 1 , . . . , fr a Gröbner basis for I . Then I = f 1 , . . . , fr .

Corollary 2.14

The proof of the proposition will use the following lemma. Let I = x α α∈A be a monomial ideal. Then every monomial in I is a multiple of some x α .

Lemma 2.15 (Key lemma)

Proof of lemma

Let x β be a monomial in I . Then we can write xβ =

x α(i) w i ,

i

where the w i are polynomials. In particular, x β appears in the right-hand side, is a monomial of x α(i) w i for some i, and thus is divisible by x α(i) .


16

Proof of proposition

We have already shown that we obtain a representation g = h 1 f 1 + · · · + h r fr

unless the algorithm stops. We need to show the algorithm terminates with gi = 0 for some i whenever g ∈ I . If g ∈ I then the intermediate gi ∈ I as well. We now use the definition of a Gröbner basis: If, for some i, the leading term LT(gi ) is not divisible by LT( f j ) for any j, then LT(gi ) ∈ LT( f 1 ), . . . , LT( fr ) by Lemma 2.15. It follows that gi ∈ I ; the formula relating g and gi guarantees that g ∈ I .

2.3

Normal forms Fix a monomial order > on k[x1 , . . . , xn ] and an ideal I ⊂ k[x1 , . . . , xn ]. Then each g ∈ k[x1 , . . . , xn ] has a unique expression g≡ cα x α (mod I ),

Theorem 2.16

x α ∈LT(I )

where cα ∈ k and all but a finite number are zero. The expression the normal form of g modulo I .

α cα x

α

is called

Equivalently, the monomials {x α : x α ∈ LT(I )} form a k-vector-space basis for the quotient k[x1 , . . . , xn ]/I . Fix a monomial order > on k[x1 , . . . , xn ], an ideal I ⊂ k[x1 , . . . , xn ], and Gröbner basis f 1 , . . . , fr for I . Then each g ∈ k[x1 , . . . , xn ] has a unique expression g≡ cα x α (mod I ),

Corollary 2.17

where LM( f j ) does not divide x α for any j or α. We first establish existence: the proof is essentially an induction on LM(g). Suppose the result is false, and consider the nonempty set

Proof of theorem:

{LM(g) : g does not admit a normal form}. One of the defining properties of monomial orders guarantees that this set has a least element x β ; choose g such that LT(g) = x β . Suppose x β ∈ LT(I ). Choose h ∈ I with LT(h) = x β and consider g˜ = g − h. ˜ < LM(g) and g˜ ≡ g (mod I ). By the minimality of g, we obtain a Note that LM(g)

2.3 NORMAL FORMS

17

normal form

g˜ ≡

cα x α

(mod I ).

x α ∈LT(I )

But this is also a normal form for g, a contradiction. ˜ < LM(g). By Now suppose x β ∈ LT(I ). Consider g˜ = g − x β so that LM(g) minimality, we have a normal form

g˜ ≡

cα x α

(mod I ).

x α ∈LT(I )

But then we have g ≡ xβ +

cα x α

(mod I ),

x α ∈LT(I )

i.e., a normal form for g, which is a contradiction. Now for uniqueness: Suppose we have g≡

α

cα x α ≡

α

c˜α x α

(mod I )

with cα = c˜α for some α. It follows that h :=

α

(cα − c˜α )x α ∈ I,

h = 0,

and LT(h) = (cα − c˜α )x α for some α. We have x α ∈ LT(I ), a contradiction. Example 2.18

Choose > such that x1 > x2 > . . . > xn .

Let 0 I k[x1 , . . . , xn ] be an ideal generated by linear forms gi =

n

ai j x j + ai0 ,

ai j ∈ k.

j=1

It is an exercise to show that gi admits a Gröbner basis of the form: f1 = b1 j x j + b10 , b1(1) = 0 j≥(1)

f2 =

b2 j x j + b20 ,

b2(2) = 0

br j x j + br 0 ,

br (r ) = 0

j≥(2)

.. . fr =

j≥(r )

18


where (1) < (2) < . . . < (r ). The numbers pivots of the row echelon form of the matrix ⎛ a11 . . . a1n ⎜ a21 . . . a2n ⎝ .. .. . ... .

(1), . . . , (r ) are positions of the ⎞ a10 a20 ⎟ ⎠. .. .

We may write {1, 2, . . . , n} = {(1), . . . , (r )} ∪ {m(1), . . . , m(n − r )}. Theorem 2.16 says that for each g ∈ k[x1 , . . . , xn ] there exists a unique P ∈ k[xm(1) , . . . , xm(n−r ) ] with g ≡ P (mod I ). Its proof implies that if g is linear we can write g ≡ c1 xm(1) + · · · + cn−r xm(n−r ) + c0

(mod I )

for unique c0 , c1 , . . . , cn−r ∈ k. Fix a monomial order > on k[x1 , . . . , xn ], a nonzero ideal I ⊂ k[x1 , . . . , xn ], and a Gröbner basis f 1 , . . . , fr for I . Given a nonzero element g ∈ k[x1 , . . . , xn ], we find the normal form of g (mod I ) as follows:

Algorithm 2.19

Put g0 = g: If each monomial appearing in g0 is not divisible by any LM( f j ) then g0 is already a normal form. Otherwise, let cβ(0) x β(0) be the largest term in g0 divisible by some LM( f j ), say LM( f j0 ). Set g1 = g0 − cβ(0) x β(0) f j0 /LT f j0

Step 0

so that g1 ≡ g0 (mod I ).. . . Given gi , if each monomial appearing in gi is not divisible by any LM( f j ) then gi is already a normal form. Otherwise, let cβ(i) x β(i) be the largest term in gi divisible by some LM( f j ), say LM( f ji ). Set Step i

gi+1 = gi − cβ(i) x β(i) f ji /LT f ji so that gi+1 ≡ gi (mod I ). The algorithm terminates in a finite number of steps, with either gi = 0 or gi in normal form.

Proposition 2.20

In passing from gi to gi+1 , we replace the largest term of gi appearing in LT(I ) with a sum of terms of lower degrees. Thus we have

Proof

x β(0) > x β(1) > . . . > x β(i) > x β(i+1) > . . . .

2.4 EXISTENCE AND CHAIN CONDITIONS

19

However, one of the defining properties of a monomial order is that every descending sequence of monomials eventually terminates, so the algorithm must terminate as well.

2.4

Existence and chain conditions We have not yet established that Gröbner bases exist, or even that each ideal of k[x1 , . . . , xn ] is finitely generated. In this section, we shall prove the following theorem. Fix a monomial order > and an arbitrary nonzero ideal I ⊂ k[x1 , . . . , xn ]. Then I admits a finite Gröbner basis for the prescribed order.

Theorem 2.21 (Existence Theorem)

We obtain the following result, named in honor of David Hilbert (1862–1943), who pioneered the use of nonconstructive arguments in algebraic geometry and invariant theory at the end of the nineteenth century: Corollary 2.22 (Hilbert Basis Theorem)

Every polynomial ideal is finitely

generated. It suffices to show that LT(I ) is finitely generated. Indeed, if f 1 , . . . , fr ∈ I are chosen such that LT(I ) = LT( f 1 ), . . . , LT( fr ) then Corollary 2.14 implies I = f 1 , . . . , fr . Thus the proof of the Existence Theorem is reduced to the case of monomial ideals: Every monomial ideal in a polynomial ring over a field is generated by a finite collection of monomials.

Proposition 2.23 (Dickson’s Lemma)

Let J ⊂ k[x1 , . . . , xn ] be a monomial ideal; we want to find a finite number of monomials {x α(1) , . . . , x α(s) } ∈ J generating J . The proof is by induction on n, the number of variables. The case n = 1 mirrors the proof in Appendix A that every ideal in k[x1 ] is principal: If x1α is the monomial of minimal degree in J and β β x1 ∈ J , then α ≤ β and x1α |x1 . For the inductive step, we assume the result is valid for k[x1 , . . . , xn ] and deduce it for k[x1 , . . . , xn , y]. Consider the following set of auxillary monomial ideals Jm ⊂ k[x1 , . . . , xn ]:

Proof

Jm = x α ∈ k[x1 , . . . , xn ] : x α y m ∈ J .


20

Note that we have an ascending chain of ideals: J0 ⊂ J1 ⊂ J2 . . . The following result will prove useful: Proposition 2.24 (Noether’s Proposition)

Let R be a ring. Then the following

conditions are equivalent: 1. 2.

every ideal I ⊂ R is finitely generated; every ascending chain of ideals I0 ⊂ I1 ⊂ I2 ⊂ . . . terminates, i.e., I N = I N +1 for sufficiently large N . Then we say the ring R is Noetherian. This terminology pays homage to Emmy Noether (1882–1935), who pioneered abstract approaches to finiteness conditions and primary decomposition. [33] Proof of Proposition 2.24

Suppose every ideal is finitely generated. Consider I ∞ = ∪n I n ,

which is also an ideal (see Exercise 2.13). Pick generators g1 , . . . , gr ∈ I∞ ; each gi ∈ Ini for some n i . If N = max(n 1 , . . . , n r ) then I∞ = I N . Conversely, suppose every ascending chain terminates. Let I be an ideal and write I = f α α∈A . If I is not generated by a finite number of α then we may construct an infinite sequence f α(1) , f α(2) , . . . with Ir := f α(1) , . . . , f α(r ) Ir +1 := f α(1) , . . . , f α(r +1) for each r , violating the ascending chain condition.

The same statement applies to S = k[x1 , . . . , xn ] with the ideals restricted to monomial ideals. Note that every monomial ideal with a finite set of generators has a finite set of monomial generators.

Remark 2.25

The sequence of monomial ideals Jm ⊂ k[x1 , . . . , xn ] terminates at some JN . Therefore, there is a finite sequence of

Completion of Proposition 2.23

2.4 EXISTENCE AND CHAIN CONDITIONS

21

monomials: x α(0,1) , . . . , x α(0,n 0 ) = J0 α(1,1) , . . . , x α(1,n 1 ) = J1 x .. . α(N ,1) α(N ,n N ) x = JN ,...,x

generating each of the Jm for m ≥ N . The ideal J is therefore generated by the terms x α(m, j) y m for m = 0, . . . , N . Essentially the same argument proves the following more general theorem: Theorem 2.26 Proof

Let R be a Noetherian ring. Then R[y] is also Noetherian.

Given J ⊂ R[y], consider

Jm = {am ∈ R : am y m + am−1 y m−1 + · · · + a0 ∈ J for some a0 , . . . , am−1 ∈ R}, i.e., the leading terms of degree m polynomials in J . We leave it to the reader to check that Jm is an ideal. Again we have an ascending sequence J0 ⊂ J1 ⊂ J2 . . . so our Noetherian assumption implies the sequence terminates at JN . Thus we can find ai j ∈ R with a0,1 , . . . , a0,n 0 = J0 a1,1 , . . . , a1,n 1 = J1 .. . a N ,1 , . . . , a N ,n N = JN Choose polynomials f i j ∈ J with leading terms ai j y i . We claim these generate J . The proof is by induction on the degree in y. Indeed, given f ∈ J we have f = bd y d + lower-order terms with bd ∈ Jd . There exist h i j ∈ R such that bd = h i j ai j . i≤d

The difference g := f − h i j f i j y d−i has degree d − 1 and is contained in J . Hence g ∈ f i j by induction and f ∈ f i j as well.


22

2.5

Buchberger’s Criterion In this section, we give an algorithm for finding a Gröbner basis for an ideal in k[x1 , . . . , xn ]. We first study how a set of generators for an ideal might fail to be a Gröbner basis. Consider I = f 1 , . . . , fr and assume that h = f 1 h 1 + · · · + fr h r

(2.2)

has leading term not contained in J = LM( f 1 ), . . . , LM( fr ). Consider the monomial x δ = max{LM( f j h j )} = max{LM( f j )LM(h j )}, j

j

which is contained in J . Therefore, the occurences of x δ in (2.2) necessarily cancel, and some smaller monomial takes up the mantle of being the leading term. We will describe precisely how such cancellation might occur: Definition 2.27

The least common multiple of monomials x α and x β is defined max(α1 ,β1 )

LCM(x α , x β ) = x1

. . . xnmax(αn ,βn ) .

Fix a monomial order on k[x1 , . . . , xn ]. Let f 1 and f 2 be polynomials in k[x1 , . . . , xn ] and set x γ (12) = LCM(LM( f 1 ), LM( f 2 )). The S-polynomial S( f 1 , f 2 ) is defined S( f 1 , f 2 ) := x γ (12) /LT( f 1 ) f 1 − x γ (12) /LT( f 2 ) f 2 . The S-polynomial is constructed to ensure the sort of cancellation alluded to above: we have x γ (12) = max LM f i x γ (12) /LT( f i ) i=1,2

but the x γ (1,2) terms cancel; in particular, LM(S( f 1 , f 2 )) < LCM(LM( f 1 ), LM( f 2 )). For example, using lexicographic order and f 1 = 2x1 x2 − x32 ,

f 2 = 3x12 − x3 ,

2.5 BUCHBERGER’S CRITERION

23

the S-polynomial is S( f 1 , f 2 ) =

x 2 x2 x12 x2 2x1 x2 − x32 − 1 2 3x12 − x3 = −1/2x1 x32 + 1/3x2 x3 . 2x1 x2 3x1

Our goal is to show that all cancellations can be expressed in terms of Spolynomials. Later on, we will put this on a more systematic footing using syzygies. Now we will prove the following: Fix a monomial order and polynomials f 1 , . . . , fr in k[x1 , . . . , xn ]. The following are equivalent:

Theorem 2.28 (Buchberger’s Criterion)

1. 2.

f 1 , . . . , fr form a Gröbner basis for f 1 , . . . , fr . Each S-polynomial S( f i , f j ) gives remainder zero on application of the division algorithm. (⇒) Each S-polynomial is contained in the ideal I = f 1 , . . . , fr . If we have a Gröbner basis, the division algorithm terminates with a representation

Proof

S( f i , f j ) =

r

h(i j)l fl

LM(S( f i , f j )) ≥ LM(h(i j)l fl )

(2.3)

l=1

In particular, S( f i , f j ) has remainder zero. (⇐) Suppose that each S-polynomial gives remainder zero; for each i, j we have an expression in the form (2.3). If the f i do not form a Gröbner basis, some h ∈ I does not have leading term in LM( f 1 ), . . . , LM( fr ). Choose a representation as in (2.2) h = h 1 f 1 + · · · + h r fr satisfying the following minimality assumptions: 1. 2.

x δ := max j {LM( f j h j )} is minimal; the number of indices j realizing the maximum (i.e., LM( f j h j ) = x δ ) is minimal. After reordering the f j , we may assume that LM( f 1 h 1 ) = LM( f 2 h 2 ) = · · · = LM( f m h m ) = x δ but LM( f j h j ) < x δ for j > m. Note that m ≥ 2 because the x δ term cancels in (2.2). For i = 1, j = 2, (2.3) takes the form S( f 1 , f 2 ) =

r

h(12)l fl ,

LM(S( f 1 , f 2 )) ≥ LM(h(12)l fl ).

l=1

Write out the S-polynomial using the definition

r x γ (12) /LT( f 1 ) f 1 − x γ (12) /LT( f 2 ) f 2 − h(12)l fl = 0 x γ (12) > LM(h(12)l fl ). l=1

(2.4)

24


Since LM( f i h i ) = x δ , i = 1, 2, we know x γ (1,2) |x δ and μx γ (12) = LT( f 1 )LT(h 1 ) for some monomial μ. We subtract μ × (2.4) from (2.2), to get a new expression h = h˜ 1 f 1 + h˜ 2 f 2 + · · · + h˜ r fr such that x δ ≥ (LM( f j h˜ j )), with strict inequality for j > m and j = 1. This contradicts the minimality assumption for (2.2). Fix a monomial order and polynomials f 1 , . . . , fr ∈ k[x1 , . . . , xn ]. A Gröbner basis for f 1 , . . . , fr is obtained by iterating the following procedure: For each i, j apply the division algorithm to the S-polynomials to get expressions

Corollary 2.29 (Buchberger’s Algorithm)

S( f i , f j ) =

r

h(i j)l fl + r (i j),

LM(S( f i , f j )) ≥ LM(h(i j)l fl )

l=1

where each LM(r (i j)) is not divisible by any of the LM( fl ). If all the remainders r (i j) = 0 then f 1 , . . . , fr are already a Gröbner basis. Otherwise, let fr +1 , . . . , fr +s denote the nonzero r (i j) and adjoin these to get a new set of generators { f 1 , . . . , fr , fr +1 , . . . , fr +s }.

Write I = f 1 , . . . , fr , S1 = { f 1 , . . . , fr }, and J1 = LM( f 1 ), . . . , LM( fr ). If J1 = LT(I ) then we are done. Otherwise, at least one of the remainders is nonzero by the Buchberger criterion. Consider S2 = { f 1 , . . . , fr , fr +1 , . . . , fr +s } and let J2 denote the ideal generated by leading terms of these polynomials. Iterating, we obtain an ascending chain of monomial ideals

Proof

J1 J2 J3 . . . ⊂ LT(I ) and subsets S1 S2 S3 . . . ⊂ I. As long as Jm LT(I ), Buchberger’s criterion guarantees that Jm Jm+1 . The chain terminates at some JN because k[x1 , . . . , xn ] is Noetherian. Since JN = JN +1 = · · · , we conclude that JN = LT(I ) and S N is a Gröbner basis for I . 2.5.1 An example

We compute a Gröbner basis of I = f 1 , f 2 = x12 − x2 , x13 − x3 with respect to lexicographic order.

2.5 BUCHBERGER’S CRITERION

25

The first S-polynomial is S( f 1 , f 2 ) = x1 f 1 − f 2 = x1 x12 − x2 − x13 − x3 = −x1 x2 + x3 ; its leading term is not contained in LM( f 1 ), LM( f 2 ) = x12 . Therefore, we must add f 3 = x1 x2 − x3 to the Gröbner basis. The next S-polynomial is S( f 1 , f 3 ) = x2 f 1 − x1 f 3 = x1 x3 − x22 ; its leading term is not contained in LM( f 1 ), LM( f 2 ), LM( f 3 ) = x12 , x1 x2 . Therefore, we must add f 4 = x1 x3 − x22 to the Gröbner basis. We have: S( f 2 , f 3 ) = x2 f 2 − x12 f 3 = x12 x3 − x2 x3 = x3 f 1 , S( f 1 , f 4 ) = x3 f 1 − x1 f 4 = x1 x22 − x2 x3 = x2 f 3 , S( f 2 , f 4 ) = x3 f 2 − x12 f 4 = x12 x22 − x32 = (x1 x2 + x3 ) f 3 , S( f 3 , f 4 ) = x3 f 3 − x2 f 4 = x23 − x32 . The last has leading term not contained in LM( f 1 ), . . . , LM( f 4 ) = x12 , x1 x2 , x1 x3 . Therefore, we must add f 5 = x23 − x32 to the Gröbner basis. Adding this new generator necessitates computing the S-polynomials involving f 5 : S( f 1 , f 5 ) = x23 f 1 − x12 f 5 = −x24 + x12 x32 = x1 x3 + x22 f 4 , S( f 2 , f 5 ) = x23 f 2 − x13 f 5 = −x23 x3 + x13 x32 = x12 x3 f 3 + x2 x3 f 1 , S( f 3 , f 5 ) = x22 f 3 − x1 f 5 = −x22 x3 + x1 x32 = x3 f 4 , S( f 4 , f 5 ) = x23 f 4 − x1 x3 f 5 = x1 x33 − x25 = x32 f 4 − x22 f 5 .


26

Buchberger’s criterion implies { f 1 , f 2 , f 3 , f 4 , f 5 } is a Gröbner basis. Note that

Remark 2.30

LM( f 2 ) ∈ LM( f 1 ), LM( f 3 ), LM( f 4 ), LM( f 5 ) so that f 2 is redundant and can be removed from the minimal Gröbner basis. The division algorithm applied to the S-polynomials for f 1 , f 3 , f 4 , f 5 gives the following relations 0= 0= 0= 0= 0= 0=

2.6

S( f 1 , S( f 1 , S( f 1 , S( f 3 , S( f 3 , S( f 4 ,

f 3 ) − f 4 = x2 f 1 − x1 f 3 − f 4 , f 4 ) − x2 f 3 = x3 f1 − x1 f 4 − x2 f 3 , f 5 ) − x1 x3 + x22 f 4 = x23 f 1 − x12 f 5 − x1 x3 + x22 f 4 , f 4 ) − f 5 = x3 f 3 − x2 f 4 − f 5 , f 5 ) − x3 f 4 = x22 f 3 − x 1 f 5 − x3 f 4 , f 5 ) − x32 f 4 + x22 f 5 = x23 − x32 f 4 − x1 x3 − x22 f 5 .

Syzygies We now formalize the notion of cancellations of leading terms of polynomials, and give an important example of how modules arise in algebraic geometry. According to the Webster Third International Unabridged Dictionary, a syzygy is the nearly straight-line configuration of three celestial bodies (as the sun, moon, and earth during a solar or lunar eclipse) in a gravitational system.

Just as the sun or moon is obscured during an eclipse, leading terms of polynomials are obscured by syzygies. The original Greek term σ υζ υγ ι´α refers to a yoke, conjunction, or copulation. Definition 2.31

Let f 1 , . . . , fr ∈ k[x1 , . . . , xn ]. A syzygy among the f j is a rela-

tion h 1 f 1 + h 2 f 2 + · · · + h r fr = 0 where (h 1 , . . . , h r ) ∈ k[x1 , . . . , xn ]r . The set of all such relations is denoted Syz( f 1 , . . . , fr ) ⊂ k[x1 , . . . , xn ]r . It is easy to check the following property of syzygies: Proposition 2.32

Syz( f 1 , . . . , fr ) is a k[x1 , . . . , xn ]-submodule of k[x1 , . . . , xn ]r .

2.6 SYZYGIES

Example 2.33

27

Given a finite set of monomials {μ j = x α( j) } j=1,...,r

with x γ (i, j) = LCM(x α(i) , x α( j) ). The syzygies among the μ j of the form x γ (i, j)−α(i) μi − x γ (i, j)−α( j) μ j generate Syz(μ1 , . . . , μr ). The proof is the same as the inductive argument for the Buchberger criterion, i.e., that all cancellations among leading terms are explained by S-polynomials (see Exercise 2.16). Let f 1 , . . . , fr be a Gröbner basis with respect to some monomial order on k[x1 , . . . , xn ]. Consider the relations γ (i j) x /LT( f i ) f i − x γ (i j) /LT( f j ) f j − h(i j)l fl = 0

Theorem 2.34

l

obtained by applying the division algorithm to the S-polynomials S( f i , f j ) = h(i j)l fl , LM(S( f i , f j )) ≥ LM(h(i j)l fl ). l

These generate Syz( f 1 , . . . , fr ) as a k[x1 , . . . , xn ]-module. Corollary 2.35 (Generalized Hilbert Basis Theorem)

The module of syzygies among

a set of polynomials is finitely generated. The proof is essentially contained in our proof of Buchberger’s criterion. Suppose we have a syzygy

Proof

f 1 h 1 + f 2 h 2 + · · · + fr h r = 0.

(2.5)

The proof proceeds by induction on x δ = max(LM( f j h j )) j

and the number of indices j realizing this maximum. After reordering, we may assume that this set of indices is {1, 2, . . . , m} with m ≥ 2. Consider the syzygy associated to f 1 and f 2 : γ (12) /LT( f 1 ) f 1 − x γ (12) /LT( f 2 ) f 2 − h(1, 2)l fl = 0. (2.6) x l

Choose the monomial μ such that μx γ (12) = LT( f 1 )LT(h 1 ). Subtract μ×(2.6) from (2.5) to get a new relation f 1 h˜ 1 + f 2 h˜ 2 + · · · + fr h˜ r = 0

28


with fewer indices realizing x δ or with max(LM( f j h˜ j )) < x δ . j

This can be placed in a much more general context: Let R be Noetherian and M ⊂ R n an R submodule. Then M is

Theorem 2.36

finitely generated. The basic strategy of the proof is induction on n; the n = 1 case is immediate because R is Noetherian. We record some elements of the argument which might be useful in other contexts:

Proof

Let M1 ⊂ M be R-modules such that M1 and M/M1 are both finitely generated. Then M is also finitely generated.

Lemma 2.37

Given generators m 1 , . . . , m s ∈ M1 and elements m s+1 , . . . , m s+t ∈ M with images generating M/M1 , m 1 , . . . , m s+t generate M. Indeed, for an arbitrary element m ∈ M, first choose rs+1 , . . . , rs+t ∈ R such that m − rs+1 m s+1 − · · · − rs+t m s+t → 0 ∈ M/M1 . But this difference is also in M1 , so we can write m − rs+1 m s+1 − · · · − rs+t m s+t = r1 m 1 + · · · + rs m s for some r1 , . . . , rs ∈ R. Iterating the previous argument gives the following result. Lemma 2.38

Suppose there exists a sequence of R-submodules 0 = M0 ⊂ M1 ⊂ M2 . . . ⊂ Mn = M

such that each Mi /Mi−1 is finitely generated. Then M is finitely generated. To prove the theorem, consider the sequence of modules 0 ⊂ R 1 ⊂ R 2 ⊂ . . . ⊂ R n−1 ⊂ R n where R j = {(r1 , r2 , . . . , r j , 0, . . . , 0) : r1 , . . . , r j ∈ R}. n− j times

Each R j ⊂ R n is a submodule and R j /R j−1 R. We have an induced sequence 0 ⊂ M1 ⊂ M2 ⊂ . . . ⊂ Mn−1 ⊂ Mn = M

2.7 EXERCISES

29

where M j = M ∩ R j . Each M j is a submodule of M. One of the standard isomorphism theorems of group theory gives M j /M j−1 = (M ∩ R j )/(M ∩ R j−1 ) → R j /R j−1 = R. In particular, we can regard each quotient as an ideal. Applying the lemmas above and the case n = 1 gives the result. The 1966 thesis [5] of Bruno Buchberger, supervised by Wolfgang Gröbner (1899–1980), introduces Gröbner bases and the Buchberger algorithm. The extension to syzygies is due to F.O. Schreyer [36]; this is presented in [9, ch. 15].

Remark 2.39

2.7

Exercises

2.1 2.2 2.3

Prove the assertion in Remark 2.2. Prove the assertion in Remark 2.12. Let I = x α(1) , . . . , x α(r ) ⊂ k[x1 , . . . , xn ] be a monomial ideal. Given a polynomial f = cβ x β ∈ I, cβ = 0, β

2.4

show that each x β is divisible by some x α( j) , j = 1, . . . , r . Let k[t1 , . . . , tn ] and fix real numbers w 1 , . . . , w n , the weights corresponding to the variables, i.e., w(t j ) = w j . Given a monomial t a with exponent a = (a1 , . . . , an ), its weight is defined w(t a ) = w 1 a1 + · · · + w n an .

2.5 2.6

2.7

We order the monomials by weight: t a > t b if and only if w(t a ) > w(t b ). This is called a weight order. (a) Take n = 2, w 1 = 3, and w 2 = 7. Is the weight order a monomial order? (b) Take n = 2, w 1 = 1, and w 2 = π . Show that the weight order is a monomial order. (c) Give necessary and sufficient conditions on the weights for the weight order to be a monomial order. (d) Can lexicographic order be defined as a weight order, in the sense defined above? (However, see [8] where a more general notion of weight order is defined.) Show there is a unique monomial order on C[x]. (a) Give an example of a monomial ideal I ⊂ C[x, y] with a minimal set of generators consisting of five elements. (b) Is there any bound on the number of generators of a monomial ideal in C[x, y]? Prove your answer! Show that x1 − x237 , x1 − x238 is not a Gröbner basis with respect to lexicographic order.


30

2.8

Consider the ideal I ⊂ C[x, y]

2.9

x 2n − y 3n , n = 1, 2, 3, 4 . . . .

Find a finite set of generators for I and show they actually generate the full ideal. Consider the polynomial f = x 4 + x 2 y 2 + y 3 − x 3 ∈ C[x, y] and the ideal I = f, ∂ f /∂ x, ∂ f /∂ y.

2.10

Compute the dimension of C[x, y]/I as a complex vector space. Determine whether x 5 ≡ y 5 (mod I ). Using the Buchberger Algorithm, compute Gröbner bases for

2.11

2.12

x3 − x15 , x2 − x13

with respect to both lexicographic order and graded reverse lexicographic order. Include all the relevant S-polynomial calculations. Which computation takes more effort? Compute the normal form of x1 x2 x3 with respect to each Gröbner basis. Fix a monomial order < on k[x1 , . . . , xn ] and a nonzero ideal I ⊂ k[x1 , . . . , xn ]. A reduced Gröbner basis for I is a Gröbner basis { f 1 , . . . , fr } with following additional properties: (1) LT( f j ) = LM( f j ) for each j, i.e., the leading coefficient of f j equals one; (2) for each i and j with i = j, no term of f i is divisible by LM( f j ). Show that I admits a unique reduced Gröbner basis. Given φ1 , . . . , φm ∈ k[x1 , . . . , xn ], consider the k-algebra homomorphism ψ : k[x1 , . . . , xn , y1 , . . . , ym ] → k[x1 , . . . , xn ] . y j → φ j xi → xi . Show that ker(ψ) = y1 − φ1 , y2 − φ2 , . . . , ym − φm . Hint: Check that the generators of I = y1 − φ1 , y2 − φ2 , . . . , ym − φm ⊂ ker(ψ) are a Gröbner basis under lexicographic order with y1 > y2 > . . . > ym > x1 > . . . > xn .

2.7 EXERCISES

31

Conclude that normal forms modulo I correspond to polynomials in k[x1 , . . . , xn ]. The inclusion j : k[x1 , . . . , xn ] → k[x1 , . . . , xn , y1 , . . . , ym ] is a right inverse for ψ, i.e., ψ ◦ j is the identity. Hence ker(ψ) ∩ k[x1 , . . . , xn ] = 0 and the quotient k[x1 , . . . , xn , y1 , . . . , ym ]/I k[x1 , . . . , xn , y1 , . . . , ym ]/ ker(ψ)

2.13

is injective. Consider an ascending chain of ideals I0 ⊂ I1 ⊂ I2 ⊂ . . .

2.14

in a ring R. Show that I∞ = ∪n In is also an ideal. Show that the ring of polynomials in an infinite number of variables k[x1 , x2 , x3 , . . . , xn , . . .]

2.15

does not satisfy the ascending chain condition. Let f 1 , f 2 ∈ k[x1 , . . . , xn ] be polynomials with no common irreducible factors. Show that Syz( f 1 , f 2 ) = ( f 2 , − f 1 )k[x1 , . . . , xn ] ⊂ k[x1 , . . . , xn ]2 .

2.16

What happens if f 1 and f 2 have common irreducible factors? Let μ j = x j for j = 1, . . . , n. Verify explicitly that Syz(μ1 , . . . , μn ) ⊂ k[x1 , . . . , xn ]n is generated by (0, . . . , 0, −x j , 0, . . . , 0, xi , 0, . . .). jth place

ith place

2.17

Consider the matrix A=

a11 a21

a12 a22

a13 a23

and the ideal I ⊂ k[a11 , a12 , a13 , a21 , a22 , a23 ] generated by the 2 × 2 minors of A, i.e., g1 = a12 a23 − a13 a22 ,

g2 = −a11 a23 + a13 a21 ,

g3 = a11 a22 − a12 a21 .


32

2.18

(a) Do the minors form a Gröbner basis with respect to lexicographic order a11 > a12 > a13 > a21 > a22 > a23 ? (b) Compute a set of generators for the syzygies among g1 , g2 , and g3 . Commutative algebra challenge problem: A ring R satisfies the descending chain condition if any descending sequence of ideals in R I1 ⊃ I2 ⊃ I3 . . . stabilizes, i.e., I N = I N +1 for large N . Such a ring is said to be Artinian. (a) Show that Z/nZ and k[t]/ t n are Artinian. (b) Show that Z and k[t] are not Artinian. (c) Show that an Artinian ring has a finite number of maximal ideals. Hint: Consider chains m1 ⊃ m1 m2 ⊃ m1 m2 m3 . . . . (d) Show that every prime ideal in an Artinian ring is maximal. Hint: Consider chains m ⊃ m2 ⊃ m3 . . . ⊃ p where p is a nonmaximal prime and m ⊃ p is maximal. (e) Show that an Artinian ring R finitely generated over a field k satisfies dimk (R) < ∞. (To say that R is finitely generated over k means that R is a quotient of a polynomial ring k[x1 , . . . , xn ].)

3 Affine varieties

In this chapter, we introduce algebraic varieties and various kinds of maps between them. Our main goal is to develop a working dictionary between geometric concepts and algebraic techniques. The geometric formulations are dictated by the algebraic structures in the background, e.g., we only consider maps that can be expressed using polynomials. For researchers in the field, geometric intuition and algebraic formalism are (or ought to be) mutually reinforcing. Most of the algebra is developed with a view toward geometric applications, rather than for its own sake. Often, complex algebraic manipulations are more transparent in vivo than in vitro. For much of the rest of this book, the base field is assumed to be infinite. Finite fields exhibit pathologies that complicate their use in algebraic geometry (cf. Definition 3.26). For instance, the polynomial function x p − x vanishes at every point of F p = Z/ pZ, where p is a prime integer. However, it does not vanish over the field extension F p2 , so we cannot identify x p − x with 0.

3.1

Ideals and varieties Definition 3.1

Given S ⊂ An (k), the ideal of polynomials vanishing on S is

defined I (S) = { f ∈ k[x1 , . . . , xn ] : f (s) = 0 for each s ∈ S}. This is an ideal: if f 1 and f 2 both vanish on S then so does f 1 + f 2 . If f vanishes on S and g is arbitrary, then g f also vanishes on S. Example 3.2

1. 2.

S = An (R) then I (S) = 0; S = {(a1 , . . . , an )} then I (S) = x1 − a1 , . . . , xn − an ; 33

34

AFF INE VARIETIES

3.

S = {(1, 1), (2, 3)} ⊂ A2 (Q) then I (S) = (x − 1)(y − 3), (x − 1)(x − 2), (y − 1)(x − 2), (y − 1)(y − 3);

4. 5.

S = N ⊂ A1 (C) then I (S) = {0}, because every nonconstant polynomial f ∈ C[x] has at most deg( f ) distinct roots (see Exercise A.13); S = {(x, y) : x 2 + y 2 = 1, x = 0} ⊂ A2 (R) has ideal I (S) = x 2 + y 2 − 1. An affine variety is the locus where a collection of polynomial equations is satisfied, i.e., given F = { f j } j∈J ⊂ k[x1 , . . . , xn ] we define

Definition 3.3

V (F) = {a ∈ An (k) : f j (a) = 0 for each j ∈ J } ⊂ An (k). These polynomials are said to define the variety. The structure of an algebraic variety on a given set depends on the choice of base field. For example, the set of rational numbers Q is a variety when it is regarded as a subset of A1 (Q), but not when it is regarded as a subset of A1 (R) or A1 (C). When we want to put particular emphasis on the ground field, we will say that V is an affine variety defined over k. This means the defining polynomials have coefficients in k. In what follows, B is a (possibly infinite) index set: For each β ∈ B, let Fβ ⊂ k[x1 , . . . , xn ] denote a collection of polynomials. Then we have

Proposition 3.4

V (∪β∈B Fβ ) = ∩β∈B V (Fβ ). Proof

V (∪β∈B Fβ ) = {a ∈ An (k) : f (a) = 0 for each f ∈ ∪β∈B Fβ } = ∩β∈B {a ∈ An (k) : f (a) = 0 for each f ∈ Fβ } = ∩β∈B V (Fβ ).

Thus as we add new polynomials to our collection, the corresponding variety gets smaller: For each collection of polynomials F = { f j } j∈J ⊂ k[x1 , . . . , xn ] and each subset F ⊂ F we have V (F ) ⊃ V (F).

Proposition 3.5

We have the analogous statements for ideals, which are left to the reader: Proposition 3.6

For each β ∈ B, let Sβ ⊂ An (k) denote a subset. Then we have I (∪β∈B Sβ ) = ∩β∈B I (Sβ ).

3.1 IDEALS AND VARIETIES

35

For any subsets S ⊂ S ⊂ An (k) we have I (S ) ⊃ I (S).

Proposition 3.7

The variety defined by a collection of polynomials only depends on the ideal they define: Given F = { f j } j∈J ⊂ k[x1 , . . . , xn ] generating an ideal I = f j j∈J , we have V (F) = V (I ).

Proposition 3.8

Proposition 3.5 guarantees V (F) ⊃ V (I ). Conversely, for v ∈ V (F) N we have f j (v) = 0 for each j ∈ J . For each g = i=1 h i f ji ∈ I we have g(v) = N h f (v) = 0, so v ∈ V (I ). i j i i=1

Proof

Given a ring R and a collection of ideals {Iβ }β∈B of R, we define

Definition 3.9

the sum to be

Iβ = { f 1 + . . . + fr : f j ∈ Iβ j for some β j },

β∈B

i.e., all finite sums of elements each taken from one of the Iβ . We leave it to the reader to check (see Exercise 3.4) that this is the smallest ideal containing ∪β∈B Iβ . Proposition 3.10

For any collection of ideals {Iβ }β∈B in k[x1 , . . . , xn ], we have Iβ = ∩β∈B V (Iβ ). V β∈B

Proof

This follows from Proposition 3.4 and the identity V Iβ = V (∪β∈B Iβ ) β∈B

of Proposition 3.8.

Given a ring R and ideals I1 , I2 ⊂ R, the product I1 I2 is the ideal generated by products f 1 f 2 with f 1 ∈ I1 and f 2 ∈ I2 .

Definition 3.11

Proposition 3.12

For any ideals I1 , I2 ⊂ k[x1 , . . . , xn ], we have V (I1 ∩ I2 ) = V (I1 I2 ) = V (I1 ) ∪ V (I2 ).

Proof

We have inclusions I1 I2 ⊂ I1 ∩ I2 ⊂ I1 , I2

36

AFF INE VARIETIES

so Proposition 3.5 yields V (I1 I2 ) ⊃ V (I1 ∩ I2 ) ⊃ V (I1 ), V (I2 ). It remains to show that V (I1 I2 ) ⊂ V (I1 ) ∪ V (I2 ). Suppose that v ∈ V (I1 I2 ) but v ∈ V (I1 ). Then for some f ∈ I1 we have f (v) = 0. However, for each g ∈ I2 , all the products ( f g)(v) = 0. Thus g(v) = 0 and v ∈ V (I2 ). We describe the behavior of varieties under set-theoretic operations: An arbitrary intersection of varieties ∩β∈B Vβ is a variety. A N finite union of varieties ∪i=1 Vi is a variety.

Proposition 3.13

For the intersection part, write each Vβ = V (Iβ ) for some ideal Iβ . Then Proposition 3.10 gives the result. To show that a finite union of varieties is a variety, we apply Proposition 3.12 successively.

Proof

The above properties and the Hilbert Basis Theorem from Chapter 2 together imply the following. Every variety can be defined as the locus where a finite number of polynomials vanish.

Proposition 3.14

We can take products of affine varieties: Definition3.15

Consider affine varieties V ⊂ An (k) and W ⊂ Am (k). The product

variety V × W ⊂ An+m (k) is defined as the set {(a1 , . . . , an , b1 , . . . , bm ) : (a1 , . . . , an ) ∈ V, (b1 , . . . , bm ) ∈ W }. The projections π1 : V × W → V,

π2 : V × W → W

are defined π1 (x1 , . . . , xn , y1 , . . . , ym ) = (x1 , . . . , xn ), = (y1 , . . . , ym ).

π2 (x1 , . . . , xn , y1 , . . . , ym )

3.1 IDEALS AND VARIETIES

37

Consider ideals I1 ⊂ k[x1 , . . . , xn ] and I2 ⊂ k[y1 , . . . , ym ] and the corresponding varieties V (I1 ) ⊂ An (k) and V (I2 ) ⊂ Am (k). Let

Proposition 3.16

J = I1 k[x1 , . . . , xn , y1 , . . . , ym ] + I2 k[x1 , . . . , xn , y1 , . . . , ym ], i.e., the ideal in k[x1 , . . . , xn , y1 , . . . , ym ] generated by I1 and I2 . Then V (I1 ) × V (I2 ) = V (J ). Proof

We start with some notation: Given a map φ : X → Y and a subset Z ⊂ Y ,

we write φ −1 (Z ) = {x ∈ X : φ(x) ∈ Z }. Consider the projection morphisms 1 : Am+n → An (x1 , . . . , xn , y1 , . . . , ym ) → (x1 , . . . , xn ), m+n 2 : A → Am (x1 , . . . , xn , y1 , . . . , ym ) → (y1 , . . . , ym ). Since −1 1 (V (I1 )) = {(a1 , . . . , an , b1 , . . . , bm ) : f (a1 , . . . , an ) = 0 for each f ∈ I1 } m+n it follows that −1 defined by I1 ⊂ k[x1 , . . . , xn ] ⊂ 1 (V (I1 )) is the variety in A k[x1 , . . . , xn , y1 , . . . , ym ]. Proposition 3.6 then implies

−1 1 (V (I1 )) = V (I1 k[x 1 , . . . , x n , y1 , . . . , ym ]). We can express the product as an intersection −1 V (I1 ) × V (I2 ) = −1 1 (V (I1 )) ∩ 2 (V (I2 )).

Proposition 3.10 then yields V (I1 ) × V (I2 ) = V (J ).

3.1.1 A warning about our definitions

We have defined an affine variety as the locus in An (k) where a collection of polynomials vanish. Over some base fields, it can be very difficult to determine precisely where a polynomial is zero! Fermat’s Last Theorem, as proven by Andrew Wiles and Richard Taylor, asserts that for any integers x, y, z with

Example 3.17

x N + yN = zN ,

N ≥ 3,

38

AFF INE VARIETIES

at least one of the three integers is zero. We may as well assume x, y, z ∈ Q; multiplying through by the least common multiple of the denominators would yield an integral solution. In our notation, Fermat’s Last Theorem takes the following form: If N ≥ 3 and V = V (x N + y N − z N ) ⊂ A3 (Q) then x yz ∈ I (V ). For any ideal I ⊂ k[x1 , . . . , xn ] we have (cf. Exercise 3.3) I (V (I )) ⊃ I. Whether equality holds is a subtle problem, depending both on the base field and the geometry of V (I ). There are many general theorems asserting that I (V ( f )) f for certain classes of polynomials f . When k = Q, a number field, or a finite field, these problems have a strong number-theoretic flavor. This area is known as Diophantine geometry or Arithmetic algebraic geometry. On the other hand, when k is algebraically closed the Nullstellensatz (Theorem 7.3) allows precise descriptions of I (V (I )) in terms of I .

3.2

Closed sets and the Zariski topology Definition 3.18

The algebro-geometric closure of a subset S ⊂ An (k) is defined

S = {a ∈ An (k) : f (a) = 0

for each f ∈ I (S)} = V (I (S)).

A subset S ⊂ An (k) is closed if S = S; U ⊂ An (k) is open if its complement An (k) \ U is closed in An (k). Example 3.19

1. 2. 3.

The closure of N ⊂ A1 (C) is the complex line A1 (C). The closure of {(x, y) : x 2 + y 2 = 1, x = 0} ⊂ A2 (R) is the circle {(x, y) : x 2 + y 2 = 1}. The open subsets of A1 (C) are the empty set and U ⊂ C with finite complement, e.g., U = C \ {a1 , . . . , ad }. You may remember open and closed sets from calculus, e.g., U ⊂ Rn is open if, for each x ∈ U , a sufficiently small ball centered at x is contained in U . There is a very general definition underlying both usages: A topological space consists of a set X and a collection of subsets Z = {Z ⊂ X } called the closed subsets of X , satisfying the following:

Definition 3.20

r ∅, X ∈ Z; r if Z 1 , Z 2 ∈ Z then Z 1 ∪ Z 2 ∈ Z; r if {Z j } j∈J ⊂ Z then ∩ j∈J Z j ∈ Z. A subset U ⊂ X is open if its complement X \ U is closed.

3.3 COORDINATE RINGS AND MORPHISMS

39

Our axioms imply that a finite intersection of open subsets is open and an arbitrary union of open subsets is open. Proposition 3.13 shows that closed subsets of affine space (in the sense of algebraic geometry) satisfy the axioms of a topological space. This is called the Zariski topology in recognition of Oscar Zariski (1899–1986). By Proposition 3.14, all Zariski open sets are of the form U = An (k) \ Z with Z closed = {a ∈ An (k) : f j (a) = 0, j = 1, . . . , r, f j ∈ k[x1 , . . . , xn ]}, i.e., where a finite set of polynomials do not simultaneously vanish. The Zariski topology on affine space induces a topology on subsets V ⊂ An (k): Z ⊂ V is closed if Z = V ∩ Y for some closed Y ⊂ An (k). If V ⊂ An (k) is an affine variety then closed subsets of V are precisely closed subsets of An (k) contained in V . This is called the Zariski topology on the affine variety. A function of topological spaces f : X → Y is continuous if for each closed Z ⊂ Y the preimage f −1 (Z ) = {x ∈ X : f (x) ∈ Z } is closed.

Definition 3.21

The concept of a ‘Zariski continuous’ function is really too weak to be of much use. For instance, any bijective function C → C is automatically Zariski continuous! One useful class of Zariski continuous functions are the morphisms introduced in Chapter 1: Let φ : An (k) → Am (k) be a morphism of affine spaces. Then φ is Zariski continuous.

Proposition 3.22

Proof

Let Z ⊂ Am (k) be closed, i.e., Z = {b ∈ Am (k) : g j (b) = 0, {g j } j∈J ⊂ k[y1 , . . . , ym ]}.

Thus φ −1 (Z ) = {a ∈ An (k) : g j (φ(a)) = 0} is also closed, because g j ◦ φ is a polynomial.

3.3

Coordinate rings and morphisms We elaborate on algebraic aspects of morphisms of affine space. Choose coordinates x1 , . . . , xn and y1 , . . . , ym on An (k) and A (k). Let φ : A (k) → Am (k) be a morphism given by the rule

Definition 3.23 m

n

φ(x1 , . . . , xn ) = (φ1 (x1 , . . . , xn ), . . . , φm (x1 , . . . , xn )),

φ j ∈ k[x1 , . . . , xn ].

40

AFF INE VARIETIES

For each f ∈ k[y1 , . . . , ym ], the pull-back by φ is defined φ ∗ f = f ◦ φ = f (φ1 (x1 , . . . , xn ), . . . , φm (x1 , . . . , xn )). We obtain a ring homomorphism φ ∗ : k[y1 , . . . , ym ] → k[x1 , . . . , xn ] y j → φ j (x1 , . . . , xn ), with the property that φ ∗ (c) = c for each constant c ∈ k, i.e., pull-back is a k-algebra homomorphism. Conversely, any k-algebra homomorphism ψ : k[y1 , . . . , ym ] → k[x1 , . . . , xn ] is determined by its values on the generators. Writing ψ j (x1 , . . . , xn ) = ψ(y j ), we obtain a morphism An (k) → Am (k) (x1 , . . . , xn ) → (ψ1 (x1 , . . . , xn ), . . . , ψm (x1 , . . . , xn )). To summarize: There is a natural correspondence between morphisms φ : A (k) → A (k) and k-algebra homomorphisms

Proposition 3.24 n

m

ψ : k[y1 , . . . , ym ] → k[x1 , . . . , xn ] identifying φ ∗ and ψ. We have already considered the ring of polynomial functions on affine space. How does this generalize to arbitrary affine varieties? Let V ⊂ An (k) be affine with ideal I (V ). We restrict polynomial functions on An (k) to V ; elements of I (V ) are zero along V , so these functions can be identified with the quotient k[x1 , . . . , xn ]/I (V ). I (V ) ⊂ k[x1 , . . . , xn ] → k[x1 , . . . , xn ]/I (V ) ↓ ↓ 0 ∈ functions on V Consider the circle V = {(x, y) : x 2 + y 2 = 1} ⊂ A2 (R) with I (V ) = x + y − 1. The polynomials x 2 and 1 − y 2 define the same function on the circle. We have

Example 3.25 2

2

x 2 ≡ 1 − y2

mod I (V ).

3.3 COORDINATE RINGS AND MORPHISMS

41

Let V ⊂ An (k) be an affine variety. The coordinate ring is defined

Definition 3.26

as the quotient ring k[V ] = k[x1 , . . . , xn ]/I (V ). Note that k[An ] = k[x1 , . . . , xn ] provided k is infinite (see Exercise 3.2). This is one point where finite fields create difficulties. Our definition of the coordinate ring requires modification in this case. ¯ φˆ : An (k) → Fix an affine variety V ⊂ An (k). Two morphisms φ, A (k) are equivalent on V if the induced pull-back homomorphisms

Definition 3.27 m

φ¯ ∗ : k[Am ] → k[V ],

φˆ ∗ : k[Am ] → k[V ]

are equal. The resulting equivalence classes are called morphisms φ : V → Am (k). Each φˆ : An (k) → Am (k) in the equivalence class is called an extension of φ to affine space. Consider the circle V = {(x, y) : x 2 + y 2 = 1} ⊂ A2 (R) with I (V ) = x + y − 1. The morphisms

Example 3.28 2

2

φ¯ : A2 (R) → A1 (R), (x, y) → x 2 , φˆ : A2 (R) → A1 (R), (x, y) → 1 − y 2 are equivalent on the circle. Equivalent morphisms are equivalent as functions: Proposition 3.29

Let V ⊂ An (k) be an affine variety and ¯ φˆ : An (k) → Am (k) φ,

¯ ˆ two morphisms equivalent on V . Then we have φ(v) = φ(v) for each v ∈ V . ¯ ˆ If φ(v) = φ(v) then they can be differentiated by coordinate functions from k[A ], i.e., we have

Proof

m

¯ ˆ yi (φ(v)) = yi (φ(v)) for some i. It follows that φ¯ ∗ yi (v) = φˆ ∗ yi (v), which violates the equivalence assumption.

42

AFF INE VARIETIES

Fix affine varieties V ⊂ An (k) and W ⊂ Am (k). A morphism φ : V → W is defined to be morphism φ : V → Am (k) with φ(V ) ⊂ W .

Definition 3.30

The geometric condition φ(V ) ⊂ W is equivalent to the algebraic condition φ ∗ I (W ) ⊂ I (V ). Indeed first assume that φ ∗ I (W ) ⊂ I (V ). Given v ∈ V and an arbitrary g ∈ I (W ), we have g(φ(v)) = φ ∗ (g)(v) = 0 as φ ∗ (g) ∈ I (V ), so that φ(v) ∈ W . Conversely, if φ(V ) ⊂ W then, given g ∈ I (W ), we have g(φ(v)) = 0 for each v ∈ V , and thus φ ∗ g ∈ I (V ). We are tacitly assuming that the polynomials defining φ have coefficients in k. If we have to make this explicit, we say that the morphism is defined over k. Let V ⊂ An (k) and W ⊂ Am (k) be affine varieties. Any morphism φ : V → W induces a k-algebra homomorphism φ ∗ : k[W ] → k[V ]. Conversely, each k-algebra homomorphism ψ : k[W ] → k[V ] can be expressed as φ ∗ for some morphism φ. Proposition 3.31

Proof

Suppose we have a morphism φ : V → W . Consider the composition φ∗

k[y1 , . . . , ym ] → k[x1 , . . . , xn ] → k[V ] ∪ ∪ ∪ I (W )

φ∗

→

I (V )

→

0

The ideal I (W ) is mapped to zero in k[V ], so there is an induced homomorphism φ ∗ : k[W ] = k[y1 , . . . , ym ]/I (W ) → k[V ]. Conversely, suppose we have a k-algebra homomorphism k[y1 , . . . , ym ] ↓ k[W ]

ψ

k[x1 , . . . , xn ] ↓ .

−→

k[V ]

It suffices to find a k-algebra homomorphism ψ : k[y1 , . . . , ym ] → k[x1 , . . . , xn ] making the diagram above commute. Indeed, Proposition 3.24 then gives a morphism φ : An (k) → Am (k) such that ψ = φ ∗ . The diagram guarantees ψ (I (W )) ⊂ I (V ), so φ (V ) ⊂ W and the induced homomorphism on coordinate rings k[W ] → k[V ] is just ψ.

3.4 RATIONAL MAPS

43

To construct ψ , consider the elements ψ(y j ) ∈ k[V ]. Lifting these to polynomials φ j ∈ k[x1 , . . . , xn ], j = 1, . . . , m, we obtain a homomorphism ψ : k[y1 , . . . , ym ] → k[x1 , . . . , xn ] . y j → φ j

making the diagram commute.

The main idea here is that polynomial rings are extraordinarily flexible; we can send the generators anywhere we like! Let V and W be affine varieties. There is a one-to-one correspondence between morphisms V → W and k-algebra homomorphisms k[W ] → k[V ].

Corollary 3.32

An isomorphism of affine varieties is a morphism φ : V → W admitting an inverse morphism φ −1 : W → V. An automorphism of an affine variety is an isomorphism φ : V → V.

Definition 3.33

One important consequence of Corollary 3.32 is that automorphisms of V correspond to k-algebra isomorphisms k[V ] → k[V ]. Example 3.34

Consider V = A2 (k) and the homomorphism ψ(x1 ) = x1 ,

ψ(x2 ) = x2 + g(x1 ),

g ∈ k[x1 ],

with inverse ψ −1 (x1 ) = x1 ,

ψ −1 (x2 ) = x2 − g(x1 ).

Each ψ = φ ∗ for some automorphism φ : A2 (k) → A2 (k). Thus each polynomial g ∈ k[x1 ] yields an automorphism of the affine plane. See [37] chapter I, section 2, esp. exercise 9, for more information. Exercise 3.14 is the classification of automorphisms of the affine line A1 (k). As far as I know, there is no intelligible classification of automorphisms of A3 (k). There is substantial research on this question: see [2], for example.

3.4

Rational maps Let k(x1 , . . . , xn ) denote the fraction field of k[x1 , . . . , xn ], consisting of quotients f /g where f, g ∈ k[x1 , . . . , xn ], g = 0. Definition 3.35

A rational map ρ : An (k) Am (k) is given by a rule

ρ(x1 , . . . , xn ) = (ρ1 (x1 , . . . , xn ), . . . , ρm (x1 , . . . , xn )),

ρ j ∈ k(x1 , . . . , xn ).

44

AFF INE VARIETIES

A rational map does not yield a well-defined function from An (k) to Am (k) – hence the dashed arrow! Represent each component ρ j as a fraction ρ j = f j /g j ,

f j , g j ∈ k[x1 , . . . , xn ];

we will generally assume that f j and g j have no common irreducible factors, for j = 1, . . . , m. Wherever any of the g j vanish, ρ is not well-defined; the closed set V ({g1 , . . . , gm }) ⊂ An (k) is called the indeterminacy locus of ρ. However, over the complement U := {(a1 , . . . , an ) ∈ An (k) : g j (a1 , . . . , an ) = 0, j = 1, . . . , m} we obtain a well-defined function to Am (k). The argument for Proposition 3.24 also yields: Each rational map ρ : An (k) Am (k) defined over k induces a k-algebra homomorphism

Proposition 3.36

ρ ∗ : k[y1 , . . . , ym ] → k(x1 , . . . , xn ), y j → ρ j (x1 , . . . , xn ). Conversely, each k-algebra homomorphism k[y1 , . . . , ym ] → k(x1 , . . . , xn ) arises from a rational map. Let W ⊂ Am (k) be an affine variety. A rational map ρ : An (k) W is a rational map ρ : An (k) Am (k) with ρ ∗ I (W ) = 0.

Definition 3.37

Example 3.38

If W = {(y1 , y2 ) : y22 = y12 + y13 } ⊂ A2 (Q) then we have the ratio-

nal map ρ : A1 W, 1 − s2 1 − s2 . s → , s2 s3 How do we define a rational map from a general affine variety ρ : V Am (k)? As in our discussion of morphisms, we realize V ⊂ An (k) as a closed subset. It is natural to define ρ as an equivalence class of rational maps ρ : An (k) Am (k), restricted to V . However, rational maps can behave badly along a variety, especially when one of the denominators of the ρ j vanishes along that variety. Example 3.39

Consider the rational map 3 ρ : A2 (R) A (R), −3 (x1 , x2 ) → x1 , x1−1 x2−1 , x2−3 ,

3.4 RATIONAL MAPS

45

which is well-defined over the open subset U = {(x1 , x2 ) : x1 x2 = 0} ⊂ A2 (R). However, ρ is not defined along the affine variety V = {(x1 , x2 ) : x1 = 0} ⊂ A2 (R). We formulate a definition to address this difficulty: Let ρ : An (k) Am (k) be a rational map with components ρ j = f j /g j with f j , g j ∈ k[x1 , . . . , xn ] having no common irreducible factors. Let V ⊂ An (k) be an affine variety with ideal I (V ) and coordinate ring k[V ]. Assume that the image of each g j in k[V ] does not divide zero. Then we say that ρ is admissible on V .

Definition 3.40

We would like an algebraic description of rational maps, generalizing Proposition 3.36 in the spirit of Corollary 3.32. We require an algebraic construction generalizing the field of fractions of a domain: Definition 3.41

The ring of fractions of a ring R is defined K = {r/s : r, s ∈ R, s not a zero divisor},

where r1 /s1 = r2 /s2 whenever r1 s2 = r2 s1 . We can realize R ⊂ K as the fractions with denominator 1 (see Exercise 3.17). For an affine variety V , let k(V ) denote the ring of fractions of the coordinate ring k[V ].

Definition 3.42

Let ρ : An (k) Am (k) be a rational map admissible on an affine variety V ⊂ An (k). Then ρ induces a k-algebra homomorphism

Proposition 3.43

ρ ∗ : k[Am ] → k(V ). Conversely, each such homomorphism arises from a suitable rational map. By hypothesis, each component function ρ j of ρ can be expressed as a fraction f j /g j , where f j , g j ∈ k[x1 , . . . , xn ] and g j does not divide zero in k[V ]. In particular, each f j /g j goes to an element of k(V ) and we obtain a homomorphism

Proof

ρ ∗ : k[y1 , . . . , ym ] → k(V ), y j → f j /g j .

46

AFF INE VARIETIES

Conversely, given such a homomorphism ψ, we can express ψ(y j ) = r j /s j with r j , s j ∈ k[V ] for j = 1, . . . , m. Choose polynomials f 1 , . . . , f m , g1 , . . . , gm ∈ k[x1 , . . . , xm ] with f j ≡ r j (mod I (V )) and g j ≡ s j (mod I (V )). The rational map with components ρ j = f j /g j induces the homomorphism ψ. With these results, we can make the general definition of a rational map: Definition 3.44

Let V ⊂ An (k) be an affine variety and ρ, ¯ ρˆ : An (k) → Am (k)

rational maps admissible on V . These are equivalent along V if the induced homomorphisms ρ¯ ∗ , ρˆ ∗ : k[Am ] → k(V ) are equal. Let V and W be affine varieties realized as closed subsets of An (k) and A (k) respectively. A rational map ρ : V W is defined as an equivalence class of rational maps ρ : An (k) W admissible on V . Each such ρ is called an extension of ρ to affine space.

Definition 3.45 m

The following analog of Corollary 3.32 is left as an exercise: Let V and W be affine varieties. There is a one-to-one correspondence between rational maps V W over k and k-algebra homomorphisms k[W ] → k(V ).

Corollary 3.46

3.5

Resolving rational maps We will introduce a systematic procedure for replacing rational maps by morphisms defined on a smaller variety. This will be used to compute the image of a rational map in Chapter 4: Let V and W be affine varieties realized in An (k) and Am (k) respectively. Consider a rational map ρ : V W obtained from a map

Proposition 3.47

An (k) An (k) (x1 , . . . , xn ) → ( f 1 /g1 , . . . , f m /gm ) admissible on V . Write g = g1 . . . gm so that ρ is well-defined over the open set U = {v ∈ V : g(v) = 0}. There is an affine variety Vg and morphisms π : Vg → V and φ : Vg → W , with the following properties:

3.5 RESOLVING RATIONAL MAPS

1. 2.

47

π (Vg ) = U ; there is a rational map ψ : V Vg , well-defined on U , such that π ◦ ψ = IdU and ψ ◦ π = IdVg ;

Vg p

f

W r

V 3.

φ = ρ ◦ π. Thus π is a birational morphism, i.e., a morphism which admits an inverse rational map. We’ll discuss these more in Chapter 6. Proof

The new affine variety is obtained by imposing invertibility by fiat: take An (k)g = {(x1 , . . . , xn , z) : zg(x1 , . . . , xn ) = 1} ⊂ An+1 (k)

so that projection π : An+1 (k) → An (k) (x1 , . . . , xn , z) → (x1 , . . . , xn ), takes An (k)g bijectively to An (k) \ {g = 0}. Similarly, define Vg = { p ∈ An (k)g : π ( p) ∈ V }. Abusing notation, we use π to designate the restriction of the projection to Vg ; it also maps Vg bijectively to U . These varieties have coordinate rings k[An (k)g ] =

k[x1 , . . . , xn , z] = k[x1 , . . . , xn ][1/g] ⊂ k(x1 , . . . , xn ) zg − 1

and k[Vg ] = k[V ][1/g] ⊂ k(V ). The rational map inverse to π is ψ(x1 , . . . , xn ) = (x1 , . . . , xn , 1/g); we use the same notation for its restriction to V . The morphism An+1 (k) → Am (k) (x1 , . . . , xn , z) → ( f 1 g2 . . . gm z, . . . , g1 . . . gi−1 f i gi+1 . . . gm z, . . .)

48

AFF INE VARIETIES

z

X = {(x, z) : xz = 1}

x

Figure 3.1

Resolving the rational map x → x1 .

restricted to An (k)g coincides with the composition π

ρ

An (k)g → An (k) Am (k) . Restricting to Vg yields φ : Vg → W with the analogous factorization.

The varieties Vg → V produced above are called affine open subsets of V . Example 3.48

The rational map ρ : A1 (k) A1 (k), 1 x → , x

is defined on the open subset U = A1 \ {0}. We can identify U with the affine variety X = A1 (k)x = {(x, z) : x z = 1}, which is a hyperbola. The morphism π is projection onto the x-axis; φ is projection onto the z-axis. Example 3.49

The rational map 2 ρ : A1 (R) A (R), x2 − 1 1 x → , , x2 + 1 x2 + 4

is defined on U = A1 (R). The rational map over the complex numbers defined by the same rule ρ : A1 (C) A2 (C) is defined on U = A1 (C) − {±i, ±2i}. Example 3.49 shows that the behavior of rational maps under field extensions can be subtle. Proposition 3.47 has content even when ρ is defined at each point of An (k),

3.5 RESOLVING RATIONAL MAPS

(0, 0, 0, 0)

V

Figure 3.2

49

W

The rational map of Example 3.50.

i.e., when g has no zeros over k. The morphism An (k)g → An (k) is bijective but is not an isomorphism unless g is constant. This reflects the possibility that g may acquire zeros over an extension of k. We will be able to say more about this once we have the Nullstellensatz at our disposal (see Proposition 8.40.) Proposition 3.47 has one major drawback. The open set U ⊂ V and the variety Vg are not intrinsic to the rational map ρ : V W . Rather they depend on the choice of extension An (k) Am (k), as shown by the following example. Consider the varieties

Example 3.50

V = {x1 = x2 = 0} ∪ {x3 = x4 = 0} ⊂ A4 (k) W = {x1 = x2 = x5 = 0} ∪ {x3 = x4 = x5 − 1 = 0} ⊂ A5 (k) with coordinate rings k[V ] = k[x1 , x2 , x3 , x4 ]/ x1 x3 , x1 x4 , x2 x3 , x2 x4 k[W ] = k[x1 , x2 , x3 , x4 , x5 ]/ x1 (x5 − 1), x2 (x5 − 1), x3 x5 , x4 x5 . The rational maps ρ : A4 (k) A5 (k) (x1 , x2 , x3 , x4 , x5 ) → (x1 , x2 , x3 , x4 , x1 /(x1 + x3 )) 4 ρ : A (k) A5 (k) (x1 , x2 , x3 , x4 , x5 ) → (x1 , x2 , x3 , x4 , x2 /(x2 + x4 )) are admissible on V and induce the same rational map ρ : V W . (Check this!) However, the corresponding open sets U = {v ∈ V : x1 + x3 = 0} and U = {v ∈ V : x2 + x4 = 0} differ, as do the affine varieties Vx1 +x3 and Vx2 +x4 .

50

AFF INE VARIETIES

How then should we define the indeterminacy locus of a general rational map of affine varieties ρ : V W ? For each extension ρ : An (k) Am (k), we have an open subset U ⊂ V where the denominators of the coordinate functions of ρ do not vanish. However, in some sense a map is well defined at all points v ∈ V for which there exists some extension well defined at v. Let ρ : V W be a rational map of affine varieties. We define the indeterminacy locus as

Provisional Definition 3.51 (Indeterminacy locus)

Z = ∩ρ (V \ U ), where the intersection is taken over all extensions ρ : An (k) Am (k) and U is the open set where ρ is well defined. Our ultimate definition will have to wait until we define the indeterminacy ideal in Chapter 8. Example 3.52

In Example 3.50 the indeterminacy locus of ρ is the origin.

The proof of Proposition 3.47 and the definition of the ‘ring of fractions’ are both instances of a very common algebraic construction:

3.5.1 Localization

Let R be a ring and S ⊂ R a multiplicative subset, i.e., for all s1 , s2 ∈ S the product s1 s2 ∈ S. The localization R[S −1 ] is defined as equivalence classes {r/s : r ∈ R, s ∈ S}, where we identify r1 /s1 ≡ r2 /s2 whenever there exists a t ∈ S with t(r1 s2 − r2 s1 ) = 0. This is a ring under the operations of addition and multiplication of fractions.

Definition 3.53

We leave it to the reader to verify that the operations are compatible with the equivalence classes (see Exercise 3.19). Example 3.54

r If R is a domain and S = R ∗ then R[S −1 ] is the field of fractions of R. r If R is a ring then the nonzero divisors form a multiplicative set S (Check this!) and r

R[S −1 ] is the ring of fractions. If R = k[x1 , . . . , xn ] and S is the multiplicative set generated by g then R[S −1 ] = k[x1 , . . . , xn ][1/g] is the coordinate ring of An (k)g , the variety introduced in the proof of Proposition 3.47.

3.6 RATIONAL AND UNIRATIONAL VARIETIES

3.6

51

Rational and unirational varieties We introduce a more flexible notion of parametrization, generalizing the regular parametrizations studied in Chapter 1. Consider an affine variety W ⊂ Am (k). A rational parametrization of W is a rational map

Definition 3.55

ρ : An (k) W such that W is the closure of the image of ρ, i.e., if U ⊂ An (k) is the open subset over which ρ is defined then ρ(U ) = W . W is unirational if it admits a rational parametrization. Here ρ is defined over k; to emphasize this, we’ll say that W is unirational over k. Here are some examples beyond those introduced in Chapter 1. Example 3.56

1.

W = {(y1 , y2 ) : y12 + y22 = 1} ⊂ A2 (Q) is unirational with parametrization ρ : A1 (Q) W 2s s2 − 1 s → , . s2 + 1 s2 + 1

2.

The image of ρ is W \ {(0, 1)}. W = {(y1 , y2 , y3 ) : y12 + y22 + y32 = 1} ⊂ A3 (Q) is unirational with parametrization ρ : A2 (Q) W, 2s 2t s2 + t 2 − 1 (s, t) → , , . s2 + t 2 + 1 s2 + t 2 + 1 s2 + t 2 + 1 The image of ρ is W \ {(0, 0, 1)}. These formulas come from stereographic projections of the unit sphere from the north pole (the points (0, 1) and (0, 0, 1) respectively). Let W be a unirational affine variety with rational parametrization ρ : An (k) W inducing

Proposition 3.57

ρ ∗ : k[W ] → k(An ) = k(x1 , . . . , xn ). Then ρ ∗ is injective and induces a field extension j : k(W ) → k(x1 , . . . , xn ).

52

AFF INE VARIETIES

Conversely, if W is an affine variety admitting an injection of k-algebras ψ : k[W ] → k(x1 , . . . , xn ) then W is unirational. We will put this on a systematic footing when we introduce dominant maps. Suppose that ρ ∗ were not injective, so there exists f = 0 ∈ k[W ] with ρ f = 0. Let U ⊂ An (k) denote the complement of the indeterminacy of ρ. We therefore have

φ(U ) ⊂ w ∈ W : f (w) = 0 W ;

Proof ∗

the strict inclusion reflects the fact that f is not identically zero on W . It follows that φ(U ) = W , a contradiction. The injection ρ ∗ allows us to regard k[W ] ⊂ k(x1 , . . . , xn ); its ring of fractions k(W ) ⊂ k(x1 , . . . , xn ) as well. Conversely, suppose we have an injection ψ as above. Corollary 3.46 yields a rational map ρ : An (k) W . If φ(U ) W then there must exist some f = 0 ∈ k[W ] vanishing on φ(U ), contradicting the assumption that ψ is injective. Example 3.58

Consider the curve

W = (y1 , y2 ) : y1m = y2n ⊂ A2 (k)

where m, n ∈ N are relatively prime. Then W is unirational via k[W ] → k(s) (y1 , y2 ) → (s n , s m ). There is a stronger notion which is also worth mentioning: An affine variety W is rational if it admits a rational parametrization ρ : An (k) W such that the induced field extension

Definition 3.59

j : k(W ) → k(x1 , . . . , xn ) is an isomorphism. We have the following algebraic characterization: Corollary 3.60

An affine variety W is rational if and only if k(W ) k(x1 , . . . , xn )

as k-algebras. Proof

The isomorphism restricts to an injective homomorphism k[W ] → k(x1 , . . . , xn ).

3.7 EXERCISES

53

Proposition 3.57 yields the rational parametrization ρ : An W . Both instances of Example 3.56 are rational. For the first case, we must check that j : Q(W ) → Q(s) is surjective. However, the fraction y1 /(1 − y2 ) goes to s. In the second case, j : Q(W ) → Q(s, t) is surjective because y1 /(1 − y3 ) goes to s and y2 /(1 − y3 ) goes to t. Example 3.58 is also rational: m and n are relatively prime so k(s m , s n ) = k(s).

Example 3.61

Here are some open problems in the field: Let f ∈ C[y1 , . . . , ym ] be an irreducible polynomial of degree d ≤ m, and V ( f ) ⊂ Am (C) the corresponding hypersurface. Is V ( f ) unirational?

Problem 3.62 (Unirationality of small degree hypersurfaces)

For general such hypersurfaces, there are no techniques for disproving unirationality. However, unirationality has been established only when d = 2, m ≥ 2, d = 3, m ≥ 3 [26], or m d [18] [34]. Indeed, m grows very rapidly as a function of d; for fixed d ≥ 4, there are many values of m for which unirationality is an open problem. On the other hand, general degree d hypersurfaces in Am (C) do not admit rational parametrizations when d > m [25] §4. For instance, the hypersurface

(y1 , y2 , y3 ) : y14 + y24 + y34 = 1

lacks one. Deciding whether hypersurfaces are rational is even more difficult: There are open problems even for cubic hypersurfaces! Given f ∈ C[x1 , . . . , x5 ] of degree 3, when is V ( f ) ⊂ A5 (C) rational? [20] It would be wonderful to have an explicit test that would decide whether C(V ) C(t1 , . . . , td )!

3.7

Exercises

3.1 3.2

Prove Propositions 3.6 and 3.7. (a) Let S ⊂ An (k). Show that I (S) = k[x1 , . . . , xn ] if and only if S = ∅. (b) Let k be an infinite field. Show that I (An (k)) = 0.

3.3

Hint: Use induction on n. For each ideal I ⊂ k[x1 , . . . , xn ], show that I (V (I )) ⊃ I.

54

AFF INE VARIETIES

3.4

Give an example where equality does not hold. Given a ring R and a collection of ideals {Iβ }β∈B of R, show that β∈B Iβ is an ideal. Prove it is the smallest ideal containing ∪β∈B Iβ : Iβ = f ∈ Iβ for some β ∈ B. β∈B

3.5

Let {Vβ }β∈B be a (possibly infinite) collection of affine varieties Vβ ⊂ An (k). Show there exist a finite number of β1 , . . . , βr ∈ B such that ∩β∈B Vβ = Vβ1 ∩ Vβ2 ∩ . . . Vβr .

3.6

3.7

3.8

(a) Show that every finite subset S ⊂ An (k) is a variety. Prove that dimk k[S] = |S|. (b) Suppose that V ⊂ An (k) is an affine variety, with |V | = ∞. Show that dimk k[V ] is not finite. Compute the Zariski closures S ⊂ A2 (Q) of the following subsets: (a) S = {(n 2 , n 3 ) : n ∈ N} ⊂ A2 (Q); (b) S = {(x, y) : x 2 + y 2 < 1} ⊂ A2 (Q); (c) S = {(x, y) : x + y ∈ Z} ⊂ A2 (Q). Let V1 , V2 ⊂ An (k) be affine varieties. Show that I (V1 ) + I (V2 ) ⊂ I (V1 ∩ V2 ).

3.9

Find an example where equality does not hold. Consider the varieties V, W ⊂ A2 (C)

V = (x1 , x2 ) : x14 + x24 = 1 ,

W = (y1 , y2 ) : y12 + y22 = 1 ,

and the morphism φ : A2 (C) → A2 (C) (x1 , x2 ) → x12 , x22 .

3.10

Show that φ(V ) ⊂ W . (Diagonal morphism) Let V be an affine variety. The diagonal map is defined :V → V ×V v → (v, v) with image V := (V ) ⊂ V × V . (a) Show that is a morphism. (b) Let V = An (k) and fix coordinates x1 , . . . , xn and y1 , . . . , yn on An (k) × An (k). Show that I ( An (k) ) = x1 − y1 , x2 − y2 , . . . , xn − yn . (c) For general V , show that V is closed in V × V and hence an affine variety.

3.7 EXERCISES

3.11

55

(d) Show that : V → V is an isomorphism. Hint: Use the projections π1 , π2 : V × V → V. Consider the following lines in affine space A3 (R):

1 = {x1 = x2 = 0},

3.12

3.13 3.14

2 = {x1 = x3 = 0},

Compute generators for I ( 1 ∪ 2 ∪ 3 ). Justify your answer. Identify 2 × 3 matrices with entries in k a11 a12 a13 A= a21 a22 a23 with the affine space A6 (k) with coordinates a11 , a12 , a13 , a21 , a22 , a23 . Show that the matrices of rank 2 are open. Write explicit polynomials vanishing along the matrices of rank ≤ 1. Show there is a one-to-one correspondence between morphisms V → A1 (k) and functions f ∈ k[V ]. Show that every automorphism of the affine line A1 (Q) takes the form x → ax + b,

3.15

3 = {x2 = x3 = 0}.

a, b ∈ Q,

a = 0.

Let Y ⊂ A2 (C) be the variety

(y1 , y2 ) : y13 = y24 .

3.16

(a) Show there is a bijective morphism φ : A1 (C) → Y. Hint: Try y1 = x 4 . (b) Show that φ is not an isomorphism, i.e., φ does not have an inverse morphism. Let W1 , W2 , and V be affine varieties. Given a morphism φ : V → W1 × W2 , show there exist unique morphisms φ1 : V → W1 and φ2 : V → W2 such that φ1 = π1 ◦ φ,

3.17

φ2 = π2 ◦ φ.

Conversely, given φ1 : V → W1 and φ2 : V → W2 , show there exists a unique morphism φ : V → W1 × W2 satisfying these identities. Let R be a ring with ring of fractions K . Show that the rule R→K r → r/1

3.18

3.19

defines an injective homomorphism R → K . Describe the rings of fractions of the following: (a) R = Z/12Z; (b) R = k[x1 , x2 ]/ x1 x2 . Let R be a ring and S ⊂ R a multiplicative subset. Show that the operations of addition and multiplication for the localization R[S −1 ] are well-defined on equivalence classes of fractions.

56

AFF INE VARIETIES

3.20 3.21

3.22

Prove Corollary 3.46. Let R be a ring, g1 , . . . , gm ∈ R, and g = g1 g2 . . . gm . Let S and T be the multiplicative sets generated by g and {g1 , . . . , gm } respectively. For example, T = {g1e1 . . . gmem : e1 , . . . , em ≥ 0}. Show that the localizations R[S −1 ] and R[T −1 ] are isomorphic. 2 (General linear group) Identify An (k) with the space of n × n matrices A = (ai j ), with coordinate ring k[a11 , . . . , a1n , a21 , . . . , a2n , . . . , an1 , . . . , ann ].

(a) Show that matrix multiplication induces a morphism 2

2

2

μ : An (k) × An (k) → An (k) (A, B) → AB. (b) Show that there is a rational map 2

2

ι : An (k) An (k) A → A−1 . Hint: Use Cramer’s Rule. 2 (c) The affine variety An (k)det(A) constructed in Proposition 3.47 is called the general linear group and denoted GLn (k). Show there are multiplication and inversion morphisms μ : GLn (k) × GLn (k) → GLn (k),

ι : GLn (k) → GLn (k).

A variety G with multiplication and inversion operations μ : G × G → G,

3.23 3.24

satisfying the axioms of a group is called an algebraic group. Verify that ρ and ρ in Example 3.50 are admissible along V and define the same rational map V W . (a) Verify the images of the maps in Example 3.56. (b) Given nonzero numbers r1 , r2 , r3 , show that

3.25

ι:G→G

y1 r1

2

+

y2 r2

2

+

y3 r3

2 =1

is unirational. Show that W = {(y1 , y2 , y3 ) : y1 y2 y3 = 1} ⊂ A3 (Q) is rational.

4 Elimination

Eliminate, eliminate, eliminate Eliminate the eliminators of elimination theory From Shreeram S. Abhyankar, Polynomials and Power Series [7, pp. 783]

Elimination theory is the systematic reduction of systems of polynomial equations in many variables to systems in a subset of these variables. For example, when a system of polynomials admits a finite number of solutions, we would like to express these as the roots of a single polynomial in one of the original variables. The language of affine varieties, morphisms, and rational maps allows us to understand elimination theory in more conceptual terms. Recall that one of our original guiding problems concerned implicitization: describe equations for the image of a morphism φ : An (k) → Am (k). In light of the theory we have developed, it makes sense to recast this in a more general context: Consider an affine variety V ⊂ An (k), with ideal I (V ), and a morphism φ : V → Am (k). Describe generators for the ideal I (φ(V )) in terms of generators for I (V ).

Problem 4.1 (Generalized Implicitization Problem)

A warning is in order: the image of a polynomial morphism is not necessarily closed, so the best we can do is to find equations for the closure of the image. We will come back to this point when we discuss projective elimination theory in Chapter 10. In this chapter, we continue to assume that the base field is infinite.

4.1

Projections and graphs We start with an example illustrating how images of morphisms can fail to be closed:

57

58

ELIMINATION

Example 4.2

Consider the variety V = {(x1 , x2 ) : x1 x2 = 1} and the morphism φ : V → A1 (k) (x1 , x2 ) → x1 .

The image φ(V ) = {x1 : x1 = 0}, which is not closed. Here φ is induced by a projection morphism A2 (k) → A1 (k); projections play an important role in elmination. Initially, we will focus on finding images of varieties under projection: Let V ⊂ Am+n (k) be an affine variety with ideal J = I (V ). Consider the projection morphism

Theorem 4.3

π : Am+n (k) → Am (k) (x1 , . . . , xn , y1 , . . . , ym ) → (y1 , . . . , ym ). Then we have π(V ) = V (J ∩ k[y1 , . . . , ym ]). Given a polynomial f ∈ k[y1 , . . . , ym ], π ∗ f is the polynomial regarded as an element in k[x1 , . . . , xn , y1 , . . . , ym ]. To establish the forward inclusion, it suffices to check that π (V ) ⊂ V (J ∩ k[y1 , . . . , ym ]). This is the case if each f ∈ J ∩ k[y1 , . . . , ym ] vanishes on π (V ). For each p = (a1 , . . . , am ) ∈ π (V ), choose q = (b1 , . . . , bn , a1 , . . . , am ) ∈ V with π(q) = p. We have

Proof

f ( p) = f (a1 , . . . , am ) = π ∗ f (b1 , . . . , bn , a1 , . . . , am ) = π ∗ f (q) = 0 as f vanishes on V . We prove the reverse inclusion V (J ∩ k[y1 , . . . , ym ]) ⊂ π (V ). Pick p = (a1 , . . . , am ) ∈ V (J ∩ k[y1 , . . . , ym ]) and f ∈ I (π (V )). Polynomials vanishing on π(V ) pull back to polynomials vanishing on V , i.e., π ∗ I (π(V )) ⊂ I (V ) = J. In particular, π ∗ f ∈ J ∩ k[y1 , . . . , ym ] so that f (a1 , . . . , am ) = 0. The key in passing to general morphisms is the graph construction: Definition 4.4

The graph φ of a morphism φ : V → W is the locus of pairs {(v, φ(v)) : v ∈ V } ⊂ V × W.

4.1 PROJECTIONS AND GRAPHS

59

There are projections π1

φ

V

π2

W

(v, φ(v))

φ(v)

v

where π1 is invertible and π2 (φ ) = φ(V ). The graph of a morphism is itself an affine variety. We will prove a more precise statement: Consider affine varieties V ⊂ An (k) and W ⊂ Am (k) and a morphism φ : V → W . Then φ is naturally an affine variety. Precisely, choose a morphism φˆ : An (k) → Am (k) extending φ with coordinate functions φ1 , . . . , φm ∈ k[x1 , . . . , xn ]. Then we have

Proposition 4.5

I (φ ) = π1∗ I (V ) + y1 − φ1 , . . . , ym − φm , V (I (φ )) = φ , where π1 : Am+n (k) → An (k) is the projection morphism. Proof

Here the projections are induced by:

(x1 , . . . , xn , y1 , . . . , ym ) (x1 , . . . , xn )

(y1 , . . . , ym )

The inclusions φ ⊂ V × W ⊂ An (k) × Am (k) = Am+n (k) yield π1∗ I (V ) ⊂ I (φ ). If v = (v 1 , . . . , v n ) ∈ V then φ(v) = (φ1 (v 1 , . . . , v n ), . . . , φm (v 1 , . . . , v n )), so y j − φ j vanishes at (v, φ(v)). This proves that I (φ ) ⊃ π1∗ I (V ) + y1 − φ1 , . . . , ym − φm . It remains to check that I (φ ) ⊂ π1∗ I (V ) + y1 − φ1 , . . . , ym − φm . For each f ∈ k[x1 , . . . , xn , y1 , . . . ym ], we have (see Exercise 2.12) f ≡ f (x1 , . . . , xn , φ1 (x1 , . . . , xn ), . . . , φm (x1 , . . . , xn )) (mod y1 − φ1 , . . . , ym − φm ),

60

ELIMINATION

i.e., each element is congruent modulo y1 − φ1 , . . . , ym − φm to a polynomial in x1 , . . . , xn . However, if f ∈ k[x1 , . . . , xn ] vanishes along φ then f ∈ I (V ). Finally, suppose we are given (v, w) ∈ V (I (φ )). We have π1∗ I (V ) ⊂ I (φ ) so that V (I (φ )) ⊂ π1−1 (V ), i.e., v ∈ V . The remaining equations y j = φ j (x1 , . . . , xn ),

j = 1, . . . , m

imply that w = φ(v), hence (v, w) ∈ φ .

A monomial order on k[x1 , . . . , xn , y1 , . . . , ym ] is an elimination order for x1 , . . . , xn if each polynomial with leading monomial in k[y1 , . . . , ym ] is actually contained in k[y1 , . . . , ym ], i.e.,

Definition 4.6

LM(g) ∈ k[y1 , . . . , ym ] ⇒ g ∈ k[y1 , . . . , ym ]. Example 4.7

1. 2.

Lexicographic order with xi > y j for each i, j is an elimination order for x1 , . . . , xn . However, this is usually relatively inefficient computationally (see Exercise 2.10). Fix monomial orders >x and > y on x1 , . . . , xn and y1 , . . . , ym respectively. Then the product order is defined as follows: We have x α y β > x γ y δ if

x α >x x γ or x α =x x γ and y β > y y δ .

3.

One fairly efficient elimination order is the product of graded reverse lexicographic orders >x and > y . Let J ⊂ k[x1 , . . . , xn , y1 , . . . , ym ] be an ideal and > an elimination order for x1 , . . . , xn . Let { f 1 , . . . , fr } be a Gröbner basis for J with respect to >. Then J ∩ k[y1 , . . . , ym ] is generated by the elements of the Gröbner basis contained in k[y1 , . . . , ym ], i.e.,

Theorem 4.8 (Elimination Theorem)

J ∩ k[y1 , . . . , ym ] = f j : f j ∈ k[y1 , . . . , ym ] ⊂ k[y1 , . . . , ym ]. It suffices to show that each element g ∈ J ∩ k[y1 , . . . , ym ] is generated by the f j ∈ k[y1 , . . . , ym ]. Choose

Proof

g ∈ (J ∩ k[y1 , . . . , ym ]) \ f j : f j ∈ k[y1 , . . . , ym ] with LM(g) minimal. Apply the division algorithm to g and let f j be a Gröbner basis element with LM( f j )|LM(g). Hence LM( f j ) ∈ k[y1 , . . . , ym ] so the definition of the elimination order implies f j ∈ k[y1 , . . . , ym ]. Thus g˜ := g − f j LT(g)/LT( f j )

4.2 IMAGES OF RATIONAL MAPS

61

˜ < is an element of (J ∩ k[y1 , . . . , ym ]) \ f j : f j ∈ k[y1 , . . . , ym ] with LM(g) LM(g), a contradiction. Example 4.9 (Solvability of varying equations)

For which values of a ∈ C is the

system x + y = a, x 2 + y2 = a3, x 3 + y3 = a5 solvable? Let Z ⊂ A3 (C) be the solution set and π : A3 (C) → A1 (C) (x, y, a) → a, projection onto the coordinate a. We want to compute the image π (Z ). We compute a Gröbner basis for the ideal I = x + y − a, x 2 + y 2 − a 3 , x 3 + y 3 − a 5 with respect to lexicographic order: x + y − a, 2y 2 − a 3 − 2ya + a 2 , 2a 5 + a 3 − 3a 4 . Thus π (Z ) is given by the solutions to the last polynomial a = 0, 1, 12 . Here are some corresponding solutions: (x, y, a) = (0, 0, 0),

4.2

(0, 1, 1),

1 1 1 , , . 4 4 2

Images of rational maps Consider a rational map ρ : An (k) Am (k) (x1 , . . . , xn ) → ( f 1 /g1 , . . . , f m /gm ), well-defined over the open set U = {g = g1 . . . gm = 0}. Proposition 3.47 yields an affine variety An (k)g and a morphism φ : An (k)g → W such that φ(An (k)g ) = ρ(U ). Recall that An (k)g ⊂ An+1 (k) is given by {(x1 , . . . , xn , z) : zg(x1 , . . . , xn ) = 1}.

62

ELIMINATION

We write down equations for the graph of φ in An+1 (k) × Am (k). Since f j /g j = g1 . . . g j−1 f j g j+1 . . . gm /g we have I (φ ) = zg − 1, y j − g1 . . . g j−1 f j g j+1 . . . gm z, j = 1, . . . , m . The equations for the image ρ(U ) are obtained by eliminating x1 , . . . , xn , z, i.e., we find generators for I (φ ) ∩ k[y1 , . . . , ym ]. Example 4.10

Compute the image of the rational map A1 (k) A2 (k) 2 x +1 1 x → . , x2 − 1 x

The associated affine variety is Y := A1 (k)x(x 2 −1) = {(x, z) : z(x 2 − 1)x − 1 = 0} and the graph of φ : Y → A2 (k) has equations I (φ ) := z(x 2 − 1)x − 1, y1 − zx(x 2 + 1), y2 − z(x 2 − 1) . Using lexicographic order, we get a Gröbner basis 2 y2 − y1 + 1 + y22 y1 , −y2 y1 + y2 + 2z, −1 + x y2 , x y1 − x − y2 y1 − y2 and the first entry is the equation for the image ρ(A1 (k)). Computing the image of a rational map from a general affine variety looks trickier – it is hard to describe the locus where the map is well-defined (cf. Example 3.50). Luckily, a complete description of the indeterminacy is not necessary. Let ρ : V W be a rational map of affine varieties V ⊂ An (k) and W ⊂ Am (k). Let Z ⊂ V be the indeterminacy locus for ρ. Choose an extension

Proposition 4.11

ρ : An (k) Am (k) , write g = g1 . . . gm , and let φ : Vg → W be the morphism given by Proposition 3.47. Then we have ρ(V \ Z ) = φ(Vg ). Recall the indeterminacy locus is the set of points in V where each extension ρ fails to be well-defined

Proof

V \ Z = ∪extensions ρ {v ∈ V : g (v) = 0} where g is the denominator of ρ .

4.2 IMAGES OF RATIONAL MAPS

63

It suffices to show that the closure of the image is independent of the choice of extension. Given ρ , ρ : An (k) Am (k) ρ

(x1 , . . . , xn ) → ( f 1 /g1 , . . . , f m /gm ), ρ

(x1 , . . . , xn ) → ( f 1 /g1 , . . . , f m /gm ) with g = g1 . . . gm , g = g1 , . . . , gm , and φ : Vg → W and φ : Vg → W the morphisms coming from Proposition 3.47, we will show that φ (Vg ) = φ (Vg ). We can also consider φ : Vg g → W , by inverting the product g g . Since φ(Vg g ) ⊂ φ (Vg ), φ (Vg ) is suffices to show that φ(Vg g ) = φ (Vg ). This is a special case of the following: Let ψ : Y → W be a morphism of varieties, h an element of k[Y ] which does not divide zero, and U = {x ∈ Y : h(x) = 0}. Then

Lemma 4.12

ψ(U ) = ψ(Y ). If ψ(U ) ψ(Y ) then there would exist an f ∈ k[W ] such that ψ ∗ f = 0 but f (ψ(u)) = 0 for each u ∈ U . But then hψ ∗ f = 0 ∈ k[Y ], contradicting the assumption that h does not divide zero.

Proof of lemma

Thus to compute the image of a rational map ρ : V W, it suffices to compute the image of any morphism φ : Vg → W given by Proposition 3.47. The graph of φ has equations I (φ ) = y j − g1 . . . g j−1 f j g j+1 . . . gm z, zg − 1 + I (V ) and equations of the image are generators for I (φ ) ∩ k[y1 , . . . , ym ].

64

ELIMINATION

Let ρ : V W be a rational map well-defined outside the closed subset Z ⊂ V . The graph of ρ is the locus Definition 4.13

{(v, ρ(v))} ⊂ (V \ Z ) × W. The graph of a rational map may not be an affine variety – this will be a crucial point when we discuss abstract varieties. Example 4.14

The graph of ρ : A2 (k) A2 (k) (x1 , x2 ) → (x1 , x2 /x1 )

satisfies the equations y1 − x1 = y2 x1 − x2 = 0. The corresponding variety contains the line x1 = x2 = y1 = 0, which lies over the indeterminacy of ρ. Let ρ : V W be a rational map with indeterminacy locus Z . The closed graph of ρ is the closure

Definition 4.15

ρ = {(v, ρ(v)) : v ∈ V \ Z } ⊂ V × W, which is an affine variety. Again, we have projections π1 : ρ → V and π2 : ρ → W . Equations of ρ can be obtained by computing the image of the rational map (Id, ρ) : V V × W . v → (v, ρ(v)). Example 4.16

Consider the rational map ρ : A1 (k) A2 (k) t → (t/(t + 1), t 2 /(t − 1))

and the induced map (Id, ρ) : A1 (k) A1 (k) × A2 (k).

4.3 SECANT VARIETIES, JOINS, AND SCROLLS

65

The corresponding morphisms are φ : A1 (k)t 2 −1 → A2 (k) (t, z) → (t(t − 1)z, t 2 (t + 1)z) 1 (π, φ) : A (k)t 2 −1 → A1 (k) × A2 (k) (t, z) → (t, t(t − 1)z, t 2 (t + 1)z) where z(t 2 − 1) = 1. The equations of the graph are given by generators of k[t, y1 , y2 ] ∩ y1 − t(t − 1)z, y2 − t 2 (t + 1)z, z(t 2 − 1) − 1 which equals

4.3

− 2y1 y2 + t − y1 + y2 , 2y12 y2 + y12 − 3y1 y2 + y2 .

Secant varieties, joins, and scrolls In this section we describe some classical geometric constructions and how elimination techniques can be applied to write down their equations. Consider the variety N = {(t1 , . . . , t N ) : t1 + t2 + · · · + t N = 1} ⊂ A N (k). For each finite set of points S = { p1 , . . . , p N } ⊂ An (k), we have a morphism σ S : N → An (t1 , . . . , t N ) → t1 p1 + · · · + t N p N , where we add the p j as vectors in k n . The image is called the affine span of S in An (k) and denoted affspan(S). We leave it to the reader to verify this is closed (cf. Exercise 4.9.) Given distinct points p1 , p2 ∈ A2 (R), affspan( p1 , p2 ) is the unique line joining them. Given distinct noncollinear points p1 , p2 , p3 ∈ A3 (R), affspan( p1 , p2 , p3 ) is the unique plane containing them.

Example 4.17

The set S = { p1 , . . . , p N } imposes independent conditions on polynomials of degree ≤ 1 if and only if σ S is injective. We say that S is in linear general position.

Proposition 4.18

Proof

σ S is not injective if there are distinct (t1 , . . . , t N ), (t1 , . . . , t N ) ∈ N with t1 p1 + · · · + t N p N = t1 p1 + · · · + t N p N .

66

ELIMINATION

Reordering indices, if necessary, we can assume t1 = t1 ; we can write p1 =

t N − t N t2 − t2 pN , p2 + · · · + t1 − t1 t1 − t1

i.e., p1 ∈ affspan({ p2 , . . . , p N }). It follows that every linear polynomial vanishing at p2 , . . . , p N also vanishes at p1 (see Exercise 4.9) so S fails to impose independent conditions on polynomials of degree ≤ 1. Conversely, suppose S fails to impose independent conditions on polynomials of degree ≤ 1. After reordering, we find I1 ( p2 , . . . , p N ) = I1 ( p1 , p2 , . . . , p N ), which implies (see Exercise 4.9) p1 ∈ affspan( p2 , . . . , p N ). We can therefore write p1 = t2 p2 + · · · + t N p N with t2 + · · · + t N = 1. In particular, σ S (1, 0, . . . , 0) = σ S (0, t2 , . . . , t N ) so σ S is not injective.

Given a variety V ⊂ An (k) and points p1 , . . . , p N ∈ V in linear general position, affspan( p1 , . . . , p N ) is called an N -secant subspace to V .

Definition 4.19

Examples include 2-secant lines and 3-secant planes. All the N -secants are contained in the image of the morphism σ N : V × . . . × V × N → An N times

(v 1 , . . . , v N , (t1 , . . . , t N )) → t1 v 1 + · · · + t N v N . The closure of the image is the N -secant variety of V Sec N (V ) = σ N (V × . . . × V × N ). Example 4.20 (Secants of twisted cubic curves)

Consider the curve V ⊂ A3 (k)

satisfying the equations

x3 − x1 x2 , x2 − x12 .

Show that Sec2 (V ) = A3 (k). We compute the image of V × V × 2 under σ2 . Let {z 1 , z 2 , z 3 } designate the coordinates on the first V and {w 1 , w 2 , w 3 } the coordinates on the second V . The homomorphism σ N∗ is induced by xi → t1 z i + t2 w i ,

i = 1, 2, 3.

4.3 SECANT VARIETIES, JOINS, AND SCROLLS

67

The defining ideal of the graph is J = z 3 − z 1 z 2 , z 2 − z 12 , w 3 − w 1 w 2 , w 2 − w 12 , t1 + t2 − 1, xi − (t1 z i + t2 w i ), i = 1, 2, 3 . Take a Gröbner basis with respect to the product of graded reverse lexicographic orders on {t1 , t2 , w 1 , w 2 , w 3 , z 1 , z 2 , z 3 } and {x1 , x2 , x3 }. This has more terms than can be produced here, but none of them involve just x1 , x2 , x3 . It follows that J ∩ k[x1 , x2 , x3 ] = 0 and Sec2 (V ) = A3 (k). The computation is a bit easier if we use the parametrization A1 (k) → V given in Example 1.5: x1 = t, x2 = t 2 , x3 = t 3 . Then we get a morphism A 1 × A 1 × 2 → A 3 (s, u, (t1 , t2 )) → (t1 s + t2 u, t1 s 2 + t2 u 2 , t1 s 3 + t2 u 3 ) with graph defined by I = x1 − (t1 s + t2 u), x2 − (t1 s 2 + t2 u 2 ), x3 − (t1 s 3 + t2 u 3 ), t1 + t2 − 1 . The Gröbner basis with respect to the product of graded reverse lexicographic orders on {t1 , t2 , s, u} and {x1 , x2 , x3 } has no terms involving x1 , x2 , x3 . Our analysis of the twisted cubic curve suggests the following variation: Definition 4.21

Let V be an affine variety and φ(1), . . . , φ(N ) : V → An

morphisms to affine space. The scroll Scroll(V ; φ(1), . . . , φ(N )) ⊂ An is defined as the closure of the image of V × N → An (v, (t1 , . . . , t N )) → t1 φ(1)(v) + · · · + t N φ(N )(v).

Example 4.22

Let V = A1 with coordinate s and consider morphisms φ(1) : A1 → A3 s → (s, 0, 0),

φ(2) : A1 → A3 s → (1, 1, s).

Equations for Scroll(V ; φ(1), φ(2)) are given by computing the intersection k[x1 , x2 , x3 ] ∩ x1 − (t1 s + t2 ), x2 − t2 , x3 − t2 s, t1 + t2 − 1 .

68

ELIMINATION

f (1)(V)

Figure 4.1

f (2)(V)

Scroll over two curves.

Compute a Gröbner basis using the product of graded reverse lexicographic orders on {t1 , t2 , s} and {x1 , x2 , x3 }:

−x3 + x1 x2 − x22 + x2 x3 , s − x1 + x2 − x3 , −x2 + t2 , x2 + t1 − 1 .

The equation is −x3 + x1 x2 − x22 + x2 x3 = 0. Let V (1), . . . , V (N ) ⊂ An be affine varieties. The join Join(V (1), . . . , V (N )) ⊂ An is defined as the closure of the image of

Definition 4.23

V (1) × V (2) . . . × V (N ) × N → An (v(1), . . . , v(N ), (t1 , . . . , t N )) → t1 v(1) + · · · + t N v(N ). Let V ⊂ An be affine and p ∈ An . The cone over V with vertex p is defined Cone(V, p) = Join(V, p).

4.4

Exercises

4.1

The cardioid is defined as the curve C ⊂ R2 with parametric representation x(θ) = cos θ + 12 cos 2θ,

4.2

y(θ ) = sin θ + 12 sin 2θ,

0 ≤ θ < 2π.

Show that C can be defined by a polynomial equation p(x, y) = 0. Hint: Introduce auxiliary variables u and v satisfying u 2 + v 2 = 1. Express x and y as polynomials in u and v; eliminate u and v to get the desired equation in x and y. Consider the image of the circle V = (x1 , x2 ) : x12 + x22 = 1 ⊂ R2

4.4 EXERCISES

69

V

p

Figure 4.2

Cone over a curve V with vertex p.

under the map φ : V → A2 (R) given by (x1 , x2 ) →

4.3

x1 x1 x2 , 2 1 + x2 1 + x22

.

Compute the equation(s) of the image. Bonus: Produce a nice graph of the real points of the lemniscate. Consider the morphism φ : A1 (C) → A2 (C) x → (x 2 , x 3 ).

4.4

Write equations for the graph of φ and compute the image of φ using elimination theory. Advice: Throw out superfluous generators from the ideal as you go along. Suppose that x, y ∈ C and satisfy the relation √ 3

x+

√

y = 1,

where we allow each of the possible roots of x and y. Show that x and y satisfy the polynomial relation x 2 − 2x − 6x y − y 3 + 3y 2 − 3y + 1 = 0.

70

ELIMINATION

Using a computer algebra system, extract the polynomial relation correponding to √ 3

4.5

x+

√

y=

√

x + y.

Hint: For the second problem, introduce auxilliary variables s, t and u with s 3 = x, t 2 = y, and u 2 = x + y. Prove the Descartes Circle Theorem: Given four mutually tangent circles C1 , . . . , C4 ⊂ R2 with radii r1 , r2 , r3 , r4 . Take ri to be negative if the other three circles are in the interior of Ci and positive otherwise. Show that

2 2 r1−2 + r2−2 + r3−2 + r4−2 = r1−1 + r2−1 + r3−1 + r4−1 . Hint: If (xi , yi ) is the center of the ith circle, the relevant equations are (xi − x j )2 + (yi − y j )2 = (ri + r j )2 .

4.6

Consider cubic polynomials p(x) = x 3 + ax 2 + bx + c over C; we regard these as an affine space with coordinates (a, b, c). We say that p has a double (resp. triple) root if there exists an α ∈ C such that (x − α)2 | p(x) (resp. (x − α)3 | p(x)).

4.7

(a) Find equations in a, b, c for the locus of cubic polynomials with a triple root. Hint: We must have a = −3α, b = 3α 2 , c = −α 3 . (b) Find equations in a, b, c for the locus of cubic polynomials with a double root. (c) Show that x 3 + x 2 + x + 1 has no multiple roots. Consider the ideal I = x1 + x2 + x3 − a, x1 + 2x2 + 4x3 − b, x1 − x2 + x3 − c and the corresponding variety V (I ) ⊂ A6 (C). Consider the projection morphism π : V (I ) → A3 (x1 , x2 , x3 , a, b, c) → (a, b, c).

4.8

Determine the image of π. Write down equations for the image and the graph of the rational maps 6 A4 (Q) A (Q) 1 1 1 1 1 1 (a) . , , , , , (x1 , x2 , x3 , x4 ) → x1 x2 x1 x3 x1 x4 x2 x3 x2 x4 x3 x4 3 ρ : A2 (k) A (k) 1 x2 1 (b) (x1 , x2 ) → , 2 . , x1 + 1 x1 + x2 x2 + 1

4.4 EXERCISES

4.9

4.10

71

For S = { p1 , . . . , p N } ⊂ An (k), show that (a) I1 (affspan(S)) = I1 (S), i.e., the polynomials of degree ≤ 1 vanishing on affspan(S) equal the polynomials of degree ≤ 1 vanishing on S; (b) affspan(S) = V (I1 (S)). Conclude affspan(S) is an affine-linear subspace (and hence a closed subset) of An (k). Let N ≤ n + 1 and consider the subset U : = {( p1 , . . . , p N ) : p1 , . . . , p N in linear general position} ⊂ An (k) × . . . × An (k) An N (k).

N times

Show that U is open and nonempty. Hint: Write each pi = (ai1 , . . . , ain ) consider the N × (n + 1) matrix ⎛

4.11

a11 ⎜ a21 ⎜ A=⎜ . ⎝ ..

a12 a22 .. .

... ... .. .

aN 1

aN 2

. . . aN n

a1n a2n .. .

⎞ 1 1⎟ ⎟ .. ⎟. .⎠ 1

Argue that p1 , . . . , p N are in linear general position if and only if some N × N minor of A has nonvanishing determinant. Let V ⊂ A4 (k) denote the image of A1 (k) → A4 (k) t → (t, t 2 , t 3 , t 4 ). (a) Extract equations for V by computing

x1 − t, x2 − t 2 , x3 − t 3 , x4 − t 4 ∩ k[x1 , x2 , x3 , x4 ].

(b) Show that V = V¯ . (c) Show that Sec2 (V ) satisfies the equation −x2 x4 + x23 − 2x1 x2 x3 + x12 x4 + x32 = 0. 4.12

(a) Consider morphisms φ(i) : A2 → A6 , i = 1, 2, 3 φ1 (s1 , s2 ) = (s1 , 0, 0, s2 , 0, 0), φ2 (s1 , s2 ) = (0, s1 , 0, 0, s2 , 0), φ3 (s1 , s2 ) = (0, 0, s1 , 0, 0, s2 ). Show that Scroll(A2 ; φ(1), φ(2), φ(3)) is given by the 2 × 2 minors of the matrix

x1 x4

x2 x5

x3 . x6

72

ELIMINATION

(b) Consider morphisms (i) : A3 → A6 , i = 1, 2 given by (1)(r1 , r2 , r3 ) = (r1 , 0, r2 , 0, r3 , 0), (2)(r1 , r2 , r3 ) = (0, r1 , 0, r2 , 0, r3 ).

4.13

Compute equations for Scroll(A3 ; (1), (2)). Consider morphisms φ(1) : A1 → A3 s → (s, 0, 0),

4.14 4.15

φ(2) : A1 → A3 s → (1, s, s 2 ).

Compute an equation for Scroll(A1 ; φ(1), φ(2). Let (1), (2) ⊂ A3 (R) be disjoint non-parallel lines. Show that Join((1), (2)) = A3 (R). Let p = (0, 0, 0) and

V = V x3 − 1, x12 + x22 + x32 − 2 . Write down an equation for Cone(V, p).

5 Resultants

In this chapter, we develop criteria for deciding whether systems of equations have solutions. These take the form of polynomials in the coefficients of the equations that vanish whenever they can be solved. The prototype is the determinant of a system of linear equations: let A = (ai j ) be an N × N matrix with entries in a field k. The system Ax = b can be solved for each b ∈ k N if and only if det(A) = 0, in which case we can put x = A−1 b. However, in general our methods will not give an explicit formula for the solution. For higher-degree equations, we allow solutions in some extension of k. Finding solutions in a given field like k = Q or F p is really a problem of number theory rather than algebraic geometry. Most of the algebraic techniques in this chapter apply over an arbitrary field. However, the geometric interpretations via elimination theory are valid only when k is infinite.

5.1

Common roots of univariate polynomials We translate the search for common solutions to a problem in linear algebra, albeit over an infinite-dimensional space: Proposition 5.1

Consider polynomials

f = am x m + am−1 x m−1 + · · · + a0 , g = bn x n + bn−1 x n−1 + · · · + b0 ∈ k[x] of degrees m and n, i.e., with am , bn = 0. The following conditions are equivalent:

r f and g have a common solution over some extension of k; r f and g share a common nonconstant factor h ∈ k[x]; r there are no polynomials A, B ∈ k[x] with A f + Bg = 1; r f, g k[x]. We prove the first two are equivalent. Suppose f and g have a common solution α ∈ L, where L/k is a field extension. Let k[α] ⊂ L denote the k-algebra

Proof

73

74

RESULTANTS

generated by α. It is a quotient q : k[x] k[α] x → α with kernel generated by a polynomial h (see Theorem A.9). Since f (α) = g(α) = 0, f, g ∈ h(x) and h is nonzero and divides f and g. But if h were constant then q would be zero, which is impossible. Conversely, suppose that f and g share a common factor h ∈ k[x]. We may assume h is irreducible, so that k[x]/h is a field. Since f and g are in the kernel of the quotient homomorphism k[x] → k[x]/h, f and g both have roots over that field. We prove the equivalence of the second and third conditions. If h| f, g then h|(A f + Bg) for any A, B, whence A f + Bg = 1. Conversely, assume A f + Bg = 1 for each A, B ∈ k[x]. Since k[x] is a principal ideal domain (PID; see §A.5) we can write f, g := h k[x] and h is a divisor of f and g. The last two conditions are equivalent because f, g = k[x] if and only 1 ∈ f, g. Example 5.2

Consider the case m = n = 1, i.e., f = a1 x + a0 ,

g = b 1 x + b0 ,

where a1 , b1 = 0. These have common roots if and only if a1 b0 − a0 b1 = 0. Example 5.3 (Geometric approach)

Consider the variety V ⊂ A5 (k) given as

V = {(x, a0 , a1 , b0 , b1 ) : a0 + a1 x = b0 + b1 x = 0} and the projection π : V → A4 (k) (x, a0 , a1 , b0 , b1 ) → (a0 , a1 , b0 , b1 ). The image π(V ) corresponds to the f and g that have common roots, as in Example 5.2. Note that the closure a a0 =0 . π(V ) = (a0 , a1 , b0 , b1 ) : det 1 b1 b0 We shall pursue this further in Section 5.3.

5.1 COMMON ROOTS OF UNIVARIATE POLYNOMIALS

Definition 5.4

75

The resultant of polynomials of positive degrees f = am x m + am−1 x m−1 + · · · + a0 , g = bn x n + bn−1 x n−1 + · · · + b0 ∈ k[x], am , bn = 0,

is defined as the (m + n) × (m + n) determinant ⎛

am ⎜ 0 ⎜ ⎜ ⎜ 0 ⎜ ⎜ ⎜ 0 Res( f, g) = det ⎜ ⎜ bn ⎜ ⎜ 0 ⎜ ⎜ ⎝ 0 0

am−1 am 0 0 bn−1 bn 0 0

··· am−1 .. .

a0 ··· .. .

··· ··· bn−1 .. .

0 b0 ··· .. .

···

0

0 a0 ··· am 0 b0 ··· bn

0 0 .. .

··· ··· .. .

am−1 0 0 .. .

··· ··· ··· .. .

bn−1

···

⎞ 0 0⎟ ⎟ ⎟ 0⎟ ⎟ ⎟ a0 ⎟ ⎟. 0⎟ ⎟ 0⎟ ⎟ ⎟ 0⎠ b0

The first n rows involve the ai s and the last m rows involve the b j ’s. This is sometimes called the Sylvester resultant, in honor of James Joseph Sylvester (1814–1897). Two nonconstant polynomials f, g ∈ k[x] have a common factor if and only if Res( f, g) = 0.

Theorem 5.5

Our first task is to explain the determinantal form of the resultant. For each d ≥ 0, let P1,d denote the polynomials in k[x] of degree ≤ d. Recall (see Exercise 1.4) that dim P1,d = d + 1 with distinguished basis {1, x, . . . , x d }. When d < 0 we define P1,d = 0. Consider the linear transformation

Proof

δ0 (d) : P1,d−m ⊕ P1,d−n → P1,d (A, B) → A f + Bg. We have image(δ0 (d)) = P1,d ∩ f, g so the following are equivalent:

r δ0 (d) is surjective for some d ≥ 0; r 1 ∈ image(δ0 (d)) for some d; r f, g = k[x]. We compute a matrix for δ0 (d). Since dim P1,d−m ⊕ P1,d−n = 2d − m − n + 2

76

RESULTANTS

this is a (d + 1) × (2d − m − n + 2) matrix. Write A = rd−m x d−m + · · · + r0 , so that A f + Bg =

d

B = sd−n x d−n + · · · + s0

x

d− j

ri1 ai2 + si1 bi2 .

i 1 +i 2 =d− j

j=0

We can represent δ0 (d)(A, B) = (rd−m , · · · , r0 , sd−n , · · · , s0 ) · ⎛

m+1 columns

am ⎜ 0 ⎜ ⎜ ⎜ 0 ⎜ ⎜ ⎜ 0 ⎜ ⎜ bn ⎜ ⎜ 0 ⎜ ⎜ ⎝ 0 0

d−m columns

··· am

a0 ··· .. .

0 a0 .. .

··· ··· ··· .. .

0 b0 ··· .. .

0 0 ··· bn 0 0

···

d−n columns

0 0 ··· 0 0 b0

··· 0 bn

0 0 .. .

··· ··· .. .

am 0 0 .. .

··· ··· ··· .. .

···

···

⎛ ⎞ ⎞ xd 0 ⎜ d−1 ⎟ ⎜x ⎟ 0⎟ ⎟⎜ . ⎟ ⎟ ⎜ .. ⎟ ⎜ ⎟ 0⎟ ⎟⎜ . ⎟ ⎟⎜ . ⎟ a0 ⎟ ⎜ . ⎟ ⎟⎜ ⎟. 0 ⎟ ⎜ .. ⎟ ⎟⎜ . ⎟ ⎜ ⎟ 0⎟ ⎟ ⎜ .. ⎟ ⎟⎜ . ⎟ ⎟ 0 ⎠⎜ ⎝ x ⎠ b0 1

n+1 columns

The linear transformation δ0 (d) is represented by a square matrix precisely when d = m + n − 1, in which case det(δ0 (m + n − 1)) = Res( f, g).

(5.1)

The next ingredient is a precise form of Proposition 5.1: Lemma 5.6

Consider nonconstant polynomials

f = am x m + am−1 x m−1 + · · · + a0 , g = bn x n + bn−1 x n−1 + · · · + b0 ∈ k[x] with am , bn = 0. The following conditions are equivalent:

r there exist polynomials A, B ∈ k[x] with A f + Bg = 1; r there exist polynomials A, B ∈ k[x], with deg(A) ≤ n − 1 and deg(B) ≤ m − 1, such that A f + Bg = 1;

r for each polynomial p ∈ k[x] of degree d, there exist C, D ∈ k[x] with deg(C) ≤ max(d − m, n − 1) and deg(D) ≤ max(d − n, m − 1), such that C f + Dg = p. The argument is quite similar to the proof of Theorem 2.34; we study possible cancellations in the expressions C f + Dg.


77

It is clear that the third condition implies the second, and the second implies the first. We prove the first implies the third. Suppose there exist A, B ∈ k[x] with A f + Bg = 1. Since ( p A) f + ( p B)g = p there exist polynomials C, D ∈ k[x] with C f + Dg = p. Choose these such that

Proof of lemma

M := max(deg(C) + m, deg(D) + n) is minimal. We claim that M ≤ max(d, m + n − 1). Assume on the contrary that M > d and M ≥ m + n. The leading terms of C f and Dg are therefore of degree > d and necessarily cancel LT(C)LT( f ) + LT(D)LT(g) = 0. This cancellation implies that deg(C) + m = deg(D) + n = M; since M ≥ m + n it follows that deg(C) ≥ n and deg(D) ≥ m. We define polynomials C = C − (LT(C)/LT(g))g, D = D − (LT(D)/LT( f )) f ∈ k[x] so that deg(C ) < deg(C), deg(D ) < deg(D), and C f + D g = C f + Dg − ( f g)(LT(C)/LT(g) + LT(D)/LT( f )) = C f + Dg = p. Since M := max(deg(C ) + m, deg(D ) + n) < M

we have a contradiction.

We rephrase this in terms of the linear transformation δ0 (d). The following are equivalent:

r 1 ∈ image(δ0 (d)) for some d; r 1 ∈ image(δ0 (m + n − 1)); r δ0 (d) is surjective for each d ≥ m + n − 1. We complete the proof of Theorem 5.5: by Proposition 5.1, f and g have a common factor if and only if there exist no A, B ∈ k[x] with A f + Bg = 1. It follows that 1 ∈ image(δ0 (m + n − 1)), δ0 (m + n − 1) is not surjective, and det(δ0 (m + n − 1)) = 0; Equation 5.1 implies Res( f, g) = 0. Conversely, if Res( f, g) = 0 then δ0 (m + n − 1) is not surjective, and the previous lemma implies that 1 ∈ image(δ0 (d)) = f, g ∩ P1,d for any d. In particular, there exist no A, B ∈ k[x] with A f + Bg = 1.

78

RESULTANTS

5.1.1 Application to discriminants

Recall that α is a root of f = am x m + am−1 xm−1 + · · · + a0 if and only if (x − α)| f ; it is a multiple root if (x − α)2 | f . A polynomial f has a multiple root if and only if f and its derivative f = mam x m−1 + (m − 1)am−1 x m−2 + · · · + 2a2 x + a1 have a common root. Indeed, if f = (x − α)e g where g(α) = 0 then f = e(x − α)e−1 g + (x − α)e g . If e ≥ 2 then f (α) = 0; if e = 1 then f (α) = 0. Therefore, a polynomial f has multiple roots only if Res( f, f ) = 0. For example, in the case m = 2 we have ⎞ ⎛ a1 a0 a2 Res( f, f ) = det ⎝ 2a2 a1 0⎠ 0 2a2 a1 = a2 (−a12 + 4a0 a2 ). In general, the leading term am occurs in each nonzero entry of the first column of the matrix computing Res( f, f ), so we have Res( f, f ) = (−1)m(m−1)/2 am disc( f ), where disc( f ) is a polynomial in a0 , · · · , am called the discriminant of f . There is disagreement as to the sign of the discriminant: some books omit the power of (−1).

Remark 5.7

5.1.2 Resultants for homogeneous polynomials

In our definition of resultants, we required that the polynomials have non-vanishing leading terms. There is an alternate approach which does not require any assumptions on the coefficients. Consider homogeneous forms in two variables F = am x0m + · · · + a0 x1m ,

G = bn x0n + · · · + b0 x1n ∈ k[x0 , x1 ]

of degrees m and n. We define Res(F, G) using the formula of Definition 5.4. We can reformulate Theorem 5.5 as follows: Let F, G ∈ k[x0 , x1 ] be nonconstant homogeneous forms. Then Res(F, G) = 0 if and only if F and G have a common nonconstant homogeneous factor.

Theorem 5.8

In particular, F and G have a nontrivial common zero over some extension of k. Here the trivial zero is (0, 0).


79

Write f (x) = F(x, 1) and g(x) = G(x, 1); these are called the dehomogenizations of F and G with respect to x1 . Suppose that F and G have a common factor H . If H is divisible by x1 then am = bn = 0 and the first column of the matrix defining Res(F, G) is zero, so the resultant vanishes. If x1 does not divide H , then h(x) = H (x, 1) is nonconstant and divides f (x) and g(x), so Res( f, g) = 0 by Theorem 5.5. Of course, Res(F, G) vanishes as well. Now suppose Res(F, G) = 0. If am = bn = 0 then x1 divides both F and G and we are done. If am = 0 and bn = 0 then

Proof

Res( f, g) = Res(F, G), so f and g have a common factor h = cd x d + · · · + c0 , cd = 0. The homogeneous polynomial H = cd x0d + cd−1 x0d−1 x1 + · · · + c0 x1d divides F and G. (This is called the homogenization of h.) We therefore focus on the case where just one of the leading coefficients is zero, e.g., am = 0 but bn = 0. Then we can express F(x0 , x1 ) = x1 E(x0 , x1 ),

E(x0 , x1 ) = am−1 x0m−1 + am−2 x0m−2 x1 + · · · + a0 x1m−1 .

Compute Res(F, G) using expansion-by-minors along the first column. The only nonzero entry in that column is bn (in the (n + 1)th row) and the corresponding minor is ⎛ ⎞ am−1 · · · a0 0 0 ··· 0 ⎜ 0 am−1 · · · a0 0 ··· 0 ⎟ ⎜ ⎟ ⎜ ⎟ .. .. .. .. ⎜ 0 . . ··· . . 0⎟ ⎜ ⎟ ⎜ ⎟ ··· 0 0 am−1 · · · a0 ⎟ ⎜ 0 ⎜ ⎟ ⎜ bn bn−1 · · · b0 0 ··· 0 ⎟ ⎜ ⎟ .. .. ⎜ ⎟ .. .. ⎝ 0 . . ··· . . 0⎠ 0 ··· 0 bn bn−1 · · · b0 which is the matrix for Res(E, G). We therefore have Res(F, G) = (−1)n bn Res(E, G), which implies Res(E, G) = 0. If am−1 = 0 then we stop; otherwise, we iterate until we obtain F(x0 , x1 ) = x1e E e (x0 , x1 ),

E e (x0 , x1 ) = am−e x0m−e + · · · + a0 x1m−e

with am−e = 0 and Res(E e , G) = 0. Then the previous argument applies, and we find a common factor H dividing E e and G.

80

RESULTANTS

5.2

The resultant as a function of the roots We continue our discussion of the resultant of two polynomials f = am x m + · · · + a0 ,

g = bn x n + · · · + b0

by finding an expression for Res( f, g) in terms of the roots of f and g. Pass to a finite extension of k over which we have factorizations (see Theorem A.17 and Exercise A.14) f = am

m

(x − αi ),

g = bn

i=1

n

(x − β j ).

j=1

The coefficients can be expressed in terms of the roots am−k = (−1)k am αi1 . . . αik , i 1 d, is algebraic over k(z 1 , . . . , z d ), and F is algebraic over k(z 1 , . . . , z d ). The remainder of the argument is analogous to a linear independence result: given linearly independent sets

Remark7.16

{z 1 , . . . , z d }, {w 1 , . . . , w e } generating the same subspace, we necessarily have d = e. The proof entails exchanging elements of the second set for elements of the first, one at a time. After reordering the z i s and w j s, we obtain a sequence of linearly independent sets {z 1 , z 2 , . . . , z d }, {w 1 , z 2 , . . . , z d }, {w 1 , w 2 , z 3 , . . . , z d }, . . . , {w 1 , w 2 , w 3 , . . . , w d } all generating the same subspace. If there were a w j with j > d it would have to be contained in the subspace generated by {w 1 , . . . , w d }. Suppose we have algebraically independent sets {z 1 , . . . , z d }, {w 1 , . . . , w e } ⊂ F such that F is algebraic over k(z 1 , . . . , z d ) and k(w 1 , . . . , w e ). We want to show that d = e. For simplicity, we assume d ≤ e. We know that w 1 is algebraic over k(z 1 , . . . , z d ), i.e., there is a nonzero f ∈ k[x1 , . . . , xd , xd+1 ] such that f (z 1 , . . . , z d , w 1 ) = 0. Since w 1 is not algebraic over k, f must involve at least one of the variables x1 , . . . , xd , say x1 . Then z 1 is algebraic over k(w 1 , z 2 , . . . , z d ), and hence F is algebraic over k(w 1 , z 2 , . . . , z d ) by Proposition A.16. We know then that w 2 is algebraic over k(w 1 , z 2 , . . . , z d ). Proposition 7.13 implies w 2 is not algebraic over k(w 1 ). We then have g = 0 ∈ k[x1 , . . . , xd , xd+1 ] such that g(w 1 , z 2 , . . . , z d , w 2 ) = 0; g must involve at least one of the variables x2 , . . . , xd , say x2 . But then z 2 is algebraic over k(w 1 , w 2 , z 3 , . . . , z d ) and F is algebraic over k(w 1 , w 2 , z 3 , . . . , z d ). We continue in this way, deducing that F is algebraic over k(w 1 , . . . , w d ). In particular, if there were a w j with j > d then this would be algebraic over k(w 1 , . . . , w d ). This contradicts our assumption on algebraic independence.

7.4

Integral elements Let R and F be Noetherian domains with R ⊂ F. An element α ∈ F is integral over R if either of the following equivalent conditions are satisfied:

1. 2.

α is a root of a monic polynomial t D + r D−1 t D−1 + · · · r0 ∈ R[t]; the R-submodule R[α] ⊂ F generated by α is finitely generated.

7.4 INTEGRAL ELEMENTS

107

The equivalence is easy: R[α] is finitely generated if and only if we can write −α D = r D−1 α D−1 + r D−2 α D−2 + · · · + r0 ,

r j ∈ R,

for some D. Assume that the fraction field L of R is contained in F. If w ∈ F is algebraic over L then there exists a nonzero p ∈ R such that pw is integral over R.

Proposition 7.17

Let g denote the irreducible polynomial for w over L. Clearing denominators if necessary, we may assume g ∈ R[t]. Write

Proof

g = a N t N + · · · + a0 ,

ai ∈ R, a N = 0,

and observe that a NN −1 g =

N

ai a NN −i−1 (ta N )i .

i=0

The polynomial N

ai a NN −i−1 u i = u N + a N −1 a N u N −1 + · · · + a0 a NN −1 ∈ R[u]

i=0

is monic and has a N w as a root. In particular, a N w is integral over R.

Let R be a unique factorization domain with fraction field L. Then every element of L integral over R is contained in R.

Proposition 7.18

Suppose α is algebraic over the fraction field L of R. Consider the evaluation homomorphism

Proof

L[t] → L(α) t → α with kernel generated by an irreducible g ∈ L[t]. Clearing denominators, we may assume g ∈ R[t] with coefficients having no common irreducible factor. If α is integral over R then there exists a monic irreducible h ∈ R[t] such that h(α) = 0. We know that g|h in L[t], so Gauss’ Lemma (Proposition A.12) implies h = f g for f ∈ R[t]. The leading coefficients of f and g are units multiplying to 1; g becomes monic after multiplying by the leading coefficient of f . Thus we may assume the irreducible polynomial for α over L is a monic polynomial g ∈ R[t]. If α ∈ L then g has degree 1, i.e., g(t) = t − α. In particular, α ∈ R.

108

NULLSTELLENSATZ

Let S ⊂ F denote the elements in F that are integral over R. Then S is a subring of F. Proposition 7.19

We recall Theorem 2.36: A submodule of a finitely generated module over a Noetherian ring is itself finitely generated. Suppose that α, β ∈ F are integral over R. Then R[α] is finitely generated over R and β remains integral over R[α], so R[α][β] is finitely generated over R[α] and R. It follows that the subrings R[α + β] and R[αβ] are also finitely generated over R and thus α + β and αβ are integral.

Proof

7.5

Proof of Nullstellensatz I Suppose m ⊂ k[x1 , . . . , xn ] is maximal. Consider the field F = k[x1 , . . . , xn ]/m, which is finitely generated as an algebra over k. By definition, a k-algebra R is finitely generated if it can be expressed as a quotient of a polynomial ring over k. Let k be an algebraically closed field. Then any field extension F/k, finitely generated as an algebra over k, is trivial.

Lemma 7.20

Assuming Lemma 7.20, we obtain the theorem: consider the induced map π1 : k[x1 ] → F. Since F is trivial over k, π1 (x1 ) = α1 for some α1 ∈ k, thus x1 − α1 ∈ m. We prove Lemma 7.20: first, observe that, if α ∈ F is algebraic over k, then it must have an irreducible polynomial f (t) ∈ k[t]. Since k is algebraically closed this is necessarily of the form t − α (see Exercise A.14), and α ∈ k. If F/k is nontrivial then Proposition 7.15 gives a nonempty subset {x j1 , . . . , x jd } ⊂ {x1 , . . . , xn } such that the images z 1 = x j1 , . . . , z d = x jd ∈ F are a transcendence basis for F over k. For notational simplicity, we reorder the x j so that z j = x j for j = 1, . . . , d. Write R = k[z 1 , . . . , z d ] ⊂ F, which is a polynomial ring because z 1 , . . . , z d are algebraically independent. Take the remaining generators for F over k w 1 = xd+1 , . . . , w n−d = xn−d , which are algebraic over L := k(z 1 , . . . , z d ). Let S ⊂ F denote the ring of elements of F integral over k[z 1 , . . . , z d ] (see Proposition 7.19). Proposition 7.17 yields nonzero p1 , . . . , pn−d ∈ k[z 1 , . . . , z d ] such that each t j := p j w j is integral over k[z 1 , . . . , z d ], i.e., t j ∈ S. Pick an element f /g ∈ k(z 1 , . . . , z d ), with f, g ∈ k[z 1 , . . . , z d ], such that g is relen−d atively prime to p1 , . . . , pn−d . It follows that p1e1 . . . pn−d ( f /g) is not in k[z 1 , . . . , z d ] for any e1 , . . . , en−d ∈ N. However, we can represent f /g ≡ q(x1 , . . . , xn ) for some

7.6 APPLICATIONS

109

polynomial q with coefficients in k, i.e., f /g = q(z 1 , . . . , z d , t1 / p1 , . . . , tn−d / pn−d ). Let e j , j = 1, . . . , n − d, denote the highest power of xd+ j appearing in q, so that en−d multiplying through by p1e1 . . . pn−d clears the denominators in q(z 1 , . . . , z d , t1 / p1 , . . . , tn−d / pn−d ). It follows that there exists a polynomial q over k such that n−d p1e1 . . . pn−d f /g = q (z 1 , . . . , z d , t1 , . . . , tn−d ) ∈ S.

e

We also have e

n−d p1e1 . . . pn−d ( f /g) ∈ k(z 1 , . . . , z d ),

thus Proposition 7.18 implies e

n−d p1e1 . . . pn−d ( f /g) ∈ k[z 1 , . . . , z d ],

a contradiction. We sketch an alternate (and much easier!) proof of Lemma 7.20 over k = C. Any finitely generated algebra over C has a countable basis, e.g., a subset of the monomials x1i1 . . . xnin . On the other hand, if x1 = α1 for any α1 ∈ C, the uncountable collection of rational functions x1 1−α ∈ F, α ∈ C are linearly independent over C.

7.6 7.6.1 When is a polynomial in the radical of an ideal?

Applications Unfortunately, computing over algebraically closed fields presents significant technical difficulties. How can a computer represent a general complex number? The Hilbert Nullstellensatz gives us a procedure for deciding whether a polynomial vanishes over the complex points of variety without ever computing over the complex numbers! We just need to check whether the polynomial is contained in the radical of the ideal generated by some defining set of equations. This can be checked over any field containing the coefficients of the polynomials at hand. We √ no longer assume that k is algebraically closed. To test whether a polynomial g ∈ I , where I = f 1 , . . . , fr , we use the following criterion. Given an ideal I = f 1 , . . . , fr ⊂ k[x1 , . . . , xn ], g ∈ only if f 1 , . . . , fr , zg − 1 = k[x1 , . . . , xn , z].

Proposition 7.21

Proof

The proof of the Hilbert Nullstellensatz gives that 1 ∈ f 1 , . . . , fr , zg − 1 ⇒ g N ∈ I for some N .

√

I if and

110

NULLSTELLENSATZ

Conversely, if g N = f 1 h 1 + · · · + fr h r then z N g N = f 1 (h 1 z N ) + · · · + fr (h r z N ) and 1 = f 1 (h 1 z N ) + · · · + fr (h r z N ) + (1 − z N g N ). Since (1 − z N g N ) = (1 − zg)(1 + zg + · · · + z N −1 g N −1 ), we conclude 1 ∈ f 1 , . . . , fr , zg − 1 . √ Gröber basis for To decide whether g ∈ I , compute a √ f 1 , . . . , fr , zg − 1 ⊂ k[x1 , . . . , xn , z]. If it contains 1 then g ∈ I .

Algorithm 7.22

√ Note, however, that we have not given an algorithm for computing generators for I for an arbitrary ideal I . 7.6.2 Irreducibility of hypersurfaces

Let k be algebraically closed. Given f ∈ k[x1 , . . . , xn ] with associated hypersurface V ( f ), we can factor f = f 1a1 . . . frar as a product of irreducibles. This yields a decomposition V ( f ) = V f 1a1 ∪ V f 2a2 . . . V frar = V ( f 1 ) ∪ V ( f 2 ) . . . V ( fr ). Are these the irreducible components of V ( f )? In other words, does the algebraic formulation of irreducibility for polynomials coincide with the geometric notion of irreducibility for varieties? Let k be algebraically closed. If f ∈ k[x1 , . . . , xn ] is irreducible as a polynomial then V ( f ) is irreducible as a variety.

Theorem 7.23

Since k[x1 , . . . , xn ] is a unique factorization domain, the ideal f is √ prime by Proposition A.11. By the Nullstellensatz, I (V ( f )) = f , so we just need √ √ to check that f = f . Given g ∈ f then g N ∈ f and f |g N . Since f is irreducible and k[x1 , . . . , xn ] is a unique factorization domain, we conclude that f |g and g ∈ f .

Proof

If f 1 , . . . , fr are irreducible and distinct in k[x1 , . . . , xn ], with k algebraically closed, then f 1a1 . . . frar = f 1 . . . fr .

Corollary 7.24

The assumption that k is algebraically closed is necessary. We have seen (Example 6.9) the irreducibility of f does not guarantee the irreducibility of V !

Remark 7.25

7.7 DIMENSION

7.7

111

Dimension Let V ⊂ An (k) be an irreducible affine variety. The dimension dim V is defined as the transcendence degree of k(V ) over k.

Definition 7.26

We outline an effective procedure to compute the dimension of a variety. Let I ⊂ k[x1 , . . . , xn ] be a prime ideal, F the quotient field of k[x1 , . . . , xn ]/I , and d the transcendence degree of F over k. By Proposition 7.15, there exist indices 1 ≤ i1 < i2 < . . . < id ≤ n such that xi1 , . . . , xid form a trascendence basis of F over k. Indeed, any maximal algebraically independent subset will do. We therefore focus on determining whether a subset of the variables is algebraically independent. For notational simplicity, we take the first few variables. The elements x1 , . . . , xe ∈ F are algebraically independent over k if and only if I ∩ k[x1 , . . . , xe ] = 0.

Proposition 7.27

The intersection can be effectively computed using the Elimination Theorem (Theorem 4.8) If x1 , . . . , xe are algebraically dependent then there exists a nonzero polynomial f ∈ k[t1 , . . . , te ] such that f (x1 , . . . , xe ) ≡ 0 (mod I ). This gives a nontrivial element of I ∩ k[x1 , . . . , xe ]. Conversely, each such element gives an algebraic dependence relation among x1 , . . . , xe .

Proof

x1 , . . . , xd ∈ F are a transcendence basis for F/k if and only if I ∩ k[x1 , . . . , xd ] = 0 and I ∩ k[x1 , . . . , xd , x j ] = 0 for each j > d.

Corollary 7.28

Nonzero elements g(x1 , . . . , xd , x j ) ∈ I ∩ k[x1 , . . . , xd , x j ] show that x j is algebraically dependent on x1 , . . . , xd . This suggests that to check whether an algebraically independent set of variables x1 , . . . , xd is a transcendence basis, we should carry out n − d distinct eliminations. We can do a bit better: Proposition 7.29

For each j = 1, . . . , n let I j = (I ∩ k[x1 , . . . , x j ])k[x1 , . . . , xn ].

Then x1 , . . . , xd form a transcendence basis for F/k if and only if 0 = I1 = I2 = . . . = Id Id+1 . . . In .

112

NULLSTELLENSATZ

These ideals can be extracted by computing a Gröbner basis with respect to a monomial order which is simultaneously an elimination order for the sets {x1 , . . . , x j }, j = d, . . . , n. Pure lexicographic order has this property. Suppose that h ∈ k[x1 , . . . , xn ] is contained in I j . By Exercise 7.13, each coefficient of h, as a polynomial in x j+1 , . . . , xn , is also contained in I j . The vanishing of I1 , . . . , Id is equivalent to the independence of x1 , . . . , xd by Proposition 7.27. Suppose that x1 . . . , xd is a transcendence basis for F/k. For each j > d there exists a nonzero g j (t1 , . . . , td , s) ∈ k[t1 , . . . , td , s] such that

Proof

g j (x1 , . . . , xd , x j ) ≡ 0

(mod I ).

If g j ∈ I j−1 then each coefficient of g j (regarded as a polynomial in s) is contained in Id , which is zero. Thus g j ∈ I j and g j ∈ I j−1 . Conversely, suppose we have the tower of ideals as described in the proposition. For j > d choose h ∈ k[x1 , . . . , xn ] ∩ k[x1 , . . . , x j ] with h ∈ I j−1 ; we may expand h=

N

h e (x1 , . . . , x j−1 )x ej

e=0

with some h e ≡ 0 (mod I ). This implies x j is algebraic over k(x1 , . . . , x j−1 ). Iterating Proposition A.16, we deduce that x j is algebraic over k(x1 , . . . , xd ).

7.8

Exercises

7.1

Let F ∈ C[x, y] be an irreducible polynomial. Consider the set V = {(x, y) ∈ C2 : F(x, y) = 0}.

7.2

Suppose that G ∈ C[x, y] is a polynomial such that G(u, v) = 0 for each (u, v) ∈ V . Show that F|G. When C is replaced by R, can we still conclude F divides G? Let V ⊂ An (k) be an affine variety with coordinate ring k[V ]. For each ideal I ⊂ k[V ], let V (I ) = {v ∈ V : g(v) = 0 for each g ∈ I }.

7.3

Assume that k is algebraically closed. Prove that V (I ) = ∅ if and only if I = k[V ]. Let I ⊂ R be an ideal. Show that the radical √ I = {g ∈ R : g N ∈ I for some N ∈ N}

7.4 7.5

is automatically an ideal. Let S ⊂ An (k) be a subset. Show that I (S) is radical. Let F/k be a finitely generated extension which is algebraic. Show this extension is finite.

7.8 EXERCISES

7.6

7.7

113

Determine the dimensions of the following varieties: (a) affine space An (k); (b) a point; (c) an irreducible hypersurface V ( f ) ⊂ An (C). Let V ⊂ A6 (Q) be defined by the two-by-two minors of the matrix

7.8

x1 x4

x2 x5

x3 . x6

We write F = Q(V ). Exhibit a transcendence base z 1 , . . . , z d for F over Q, and express F explicitly as an algebraic extension over Q(z 1 , . . . , z d ). Let F be the quotient field of C[x1 , x2 , x3 , x4 , x5 ]/ x12 + x22 + x32 − 1 .

7.9 7.10

Exhibit a transcendence base z 1 , . . . , z d for F over C, and express F explicitly as an algebraic extension over C(z 1 , . . . , z d ). Prove Proposition 7.13. Let R be a domain with fraction field L, and assume that α is algebraic over L. Show that {r ∈ R : r α integral over R}

7.11 7.12

is a nonzero ideal in R. Let V → W be a dominant morphism of affine varieties. Show that dim V ≥ dim W . Let R and S be Noetherian integral domains with R ⊂ S. Suppose that α, β ∈ S are roots of the monic polynomials x 2 + a1 x + a0 , x 2 + b1 x + b0 ∈ R[x]

7.13

respectively. Using Gröbner basis techniques, exhibit a monic polynomial that has α + β as a root. Do the same for αβ. Let I ⊂ k[x1 , . . . , xn ] be an ideal and set I j = (I ∩ k[x1 , . . . , x j ])k[x1 , . . . , xn ] so that I1 ⊂ I2 ⊂ . . . ⊂ In . Given h ∈ k[x1 , . . . , xn ] write h=

α j+1 ...αn

α

j+1 cα j+1 ...αn x j+1 . . . xnαn

such that each cα j+1 ...αn ∈ k[x1 , . . . , x j ]. Show that if h ∈ I j then each cα j+1 ...αn ∈ I j .

114

NULLSTELLENSATZ

7.14

Given a monomial x α = x1α1 . . . xnαn , write

rad(x α ) =

xi .

i with αi =0

For each monomial ideal I = x α(i) = x1α(i,1) . . . xnα(i,n) i=1,... ,r show that √

7.15

7.16 7.17

I = rad(x α(i) ) i=1,... ,r .

Assume that the generating set of monomials for I is minimal, i.e., given distinct i, j = 1, · · · r , x α(i) does not divide x α( j) . Show that I is radical if and only if each α(i, ) = 0 or 1. Write down all the radical monomial ideals in k[x1 , x2 , x3 ] and describe the corresponding varieties in A3 (k). Show that every maximal ideal m ⊂ R[x1 , x2 ] is one of the following: r m = x1 − α1 , x2 − α2 for some α1 , α2 ∈ R; r m = x2 − r x1 − s, x 2 + bx1 + c for some r, s, b, c ∈ R with b2 − 4c < 0; 1 r m = x1 − t, x 2 + bx2 + c for some t, b, c ∈ R with b2 − 4c < 0. 2 Let I = f 1 , f 2 ⊂ C[x, y] be an ideal√generated by a linear and an irreducible quadratic polynomial. Suppose that g ∈ I . Show that g 2 ∈ I . (Geometric version of Proposition 7.27) Let V ⊂ An (k) be an irreducible affine variety over an infinite field. Show that dim(V ) ≥ d if and only if there exists a subset {x j1 , . . . , x jd } ⊂ {x1 , . . . , xn } so the projection morphism π : An (k) → Ad (k) (x1 , . . . , xn ) → (x j1 , . . . , x jd )

7.18

is dominant. Let I and J be ideals in a ring R. (a) Show that √

I∩J=

√

I∩

√

J.

(b) On the other hand, give an example in R = k[x1 , . . . , xn ] where √ √ √ I + J = I + J . 7.19

(Codimension-1 varieties are hypersurfaces) Show that any irreducible variety V ⊂ An (k) of dimension n − 1 is a hypersurface. Hint: It suffices to prove that any prime ideal I ⊂ k[x1 , . . . , xn ] with I ∩ k[x1 , . . . , xn−1 ] = 0 is principal.

7.8 EXERCISES

115

Assuming I = 0, we can express I = f 1 , . . . , fr where each f i is irreducible and does not divide any of the f j , j = i. Suppose r > 1 and consider f 1 , f 2 in L[xn ] where L = k(x1 , . . . , xn−1 ). The resultant Res( f 1 , f 2 ) is defined; verify that Res( f 1 , f 2 ) ∈ k[x1 , . . . , xn−1 ] ∩ I and thus is zero. Deduce the existence of an irreducible polynomial in L[xn ] dividing both f 1 and f 2 . On clearing denominators and applying Gauss’ Lemma, we obtain an irreducible h ∈ k[x1 , . . . , xn ] with h| f 1 , f 2 .

8 Primary decomposition

We have shown that every variety is a union of irreducible subvarieties V = V1 ∪ V2 ∪ . . . ∪ Vr ,

Vi ⊂ V j , i = j.

Our goal here is to find an algebraic analog of this decomposition, applicable to ideals. Let I = f ⊂ k[x1 , . . . , xn ] be principal and decompose the generator into irreducible elements

Example 8.1

f = f 1e1 . . . frer , where no two of the f j are proportional. Then we can write I = f 1e1 ∩ f 2e2 . . . frer = P1e1 ∩ . . . ∩ Prer ,

Pi = f i .

Note that Pi is prime by Proposition A.11. A warning is in order: decomposing even a univariate polynomial into irreducible components can be tricky in practice; the decomposition is very sensitive to the base field k. For example, given a finite extension L/Q and f ∈ Q[t], factoring f into irreducible polynomials over L is really a number-theoretic problem rather than a geometric one. This makes it challenging to implement primary decomposition on a computer, although there are algorithms for extracting some information about the decomposition [10].

8.1

Irreducible ideals Here we emulate the decomposition into irreducible components described in Theorem 6.4. Keep in mind that unions of varieties correspond to intersections of ideals (cf. Propositions 3.6 and 3.12).

116

8.1 IRREDUCIBLE IDEALS

117

An ideal I ⊂ R is reducible if can be expressed as the intersection of two larger ideals in R

Definition 8.2

I = J1 ∩ J2 ,

I J1 , J2 .

An ideal is irreducible if it is not reducible. Let R be Noetherian. Then any ideal I ⊂ R can be written as an intersection of irreducible ideals

Proposition 8.3

I = I1 ∩ I2 . . . ∩ Im . We say that the decomposition is weakly irredundant if none of the I j can be left out, i.e., I j ⊃ I1 ∩ . . . I j−1 ∩ I j+1 . . . ∩ Ir . Suppose this is not the case, so we get an infinite sequence of decompositions I = I [1] ∩ I [1] , with I [1], I [1] I , I [1] = I [2] ∩ I [2], with I [2], I [2] I [1], etc. Thus we obtain an infinite ascending sequence of ideals

Proof

I I [1] I [2]. . . , violating the fact that R is Noetherian.

A variety V is irreducible precisely when I (V ) is prime (Theorem 6.5). This connection persists in our algebraic formulation: Proposition 8.4

Any prime ideal is irreducible.

Suppose that I is prime and I = J1 ∩ J2 with I J1 . Pick f ∈ J1 with f ∈ I . Given g ∈ J2 we have

Proof

f g ∈ J1 J2 ⊂ J1 ∩ J2 = I. Since I is prime, it follows that g ∈ I . We conclude that I = J2 . Example 8.5

If I = x y, y 2 ⊂ k[x, y] then we can write I = y ∩ x 2 , y.

The first term is prime, hence irreducible, so we focus on the second term. Suppose there were a decomposition x 2 , y = J1 ∩ J2

118

PRIMARY DECOMPOSITION

with I J1 , J2 . The quotient ring R = k[x, y]/x 2 , y has dimension dimk R = 2, so dimk k[x, y]/Ji = 1 and each Ji is a maximal ideal containing x 2 , y. The only possibility is x, y, which contradicts our assumptions. Theorem 6.4 also asserts the uniqueness of the decomposition into irreducible components. This is conspicuously lacking from Proposition 8.3. Unfortunately, uniqueness fails in the algebraic situation: Example 8.6

Consider I = x 2 , x y, y 2 ⊂ k[x, y]. We have I = y, x 2 ∩ y 2 , x = y + x, x 2 ∩ x, (y + x)2 .

The analysis in Example 8.5 shows that the ideals involved are all irreducible. For the rest of this chapter, we will develop partial uniqueness results for representations of ideals as intersections of irreducible ideals.

8.2

Quotient ideals Definition 8.7

Given ideals I, J in a ring R, the quotient ideal is defined I : J = {r ∈ R : r s ∈ I for each s ∈ J }.

For any ideals I, J, K ⊂ R, I J ⊂ K if and only if I ⊂ K : J . This explains our choice of terminology. Proposition 8.8

1. 2. 3.

For ideals I, J ⊂ k[x1 , . . . , xn ] we have:

I : J ⊂ I (V (I ) \ V (J )); V (I : J ) ⊃ V (I ) \ V (J ); I (V ) : I (W ) = I (V \ W ). Given f ∈ I : J , we have f g ∈ I for each g ∈ J . For x ∈ V (I ) \ V (J ), there exists a g ∈ J such that g(x) = 0. Since ( f g)(x) = 0 we have f (x) = 0, which proves the first assertion. The second assertion follows by taking varieties associated to the ideals in the first assertion. The inclusion I (V ) : I (W ) ⊂ I (V \ W ) follows from the first assertion; set I = I (V ) and J = I (W ). To prove the reverse inclusion, note that if f ∈ I (V \ W ) and g ∈ I (W ) then f g ∈ I (V ) and thus f ∈ I (V ) : I (W ).

Proof

We develop algorithms for computing the quotient by reducing its computation to the computation of intersections: Proposition 8.9

Let I ⊂ R be an ideal and g ∈ k[x1 , . . . , xn ]. If I ∩ g = h 1 , . . . , h s

8.3 PRIMARY IDEALS

119

then I : g = h 1 /g, h 2 /g, . . . , h s /g. Given an ideal J = g1 , . . . , gr we have I : J = (I : g1 ) ∩ (I : g2 ) ∩ . . . ∩ (I : gr ). Each h i /g ∈ I : g, so the inclusion ⊃ is clear. Given f ∈ I : g we have g f ∈ I, g, i.e., g f ∈ I ∩ g. The last assertion is an immediate consequence of the definitions.

Proof

To compute the quotient I : J of ideals I, J ⊂ k[x1 , . . . , xn ] with J = g1 , . . . , gr , we carry out the following steps:

Algorithm 8.10

1. 2. 3.

compute the intersections I ∩ gi using Proposition 6.19; using the first part of Proposition 8.9, write out generators for I : gi in terms of the generators of I ∩ gi ; compute the intersection (I : g1 ) ∩ (I : g2 ) ∩ . . . ∩ (I : gr ), which is the desired quotient I : J . Remark 8.11

To compute the intersection of a finite collection of ideals J1 , . . . , Jm ⊂ k[x1 , . . . , xn ],

the following formula is useful: J1 ∩ . . . ∩ Jm = (s1 J1 + · · · + sm Jm + s1 + · · · + sm − 1) ∩ k[x1 , x2 , . . . , xn ].

8.3

Primary ideals An ideal I in a ring R is primary if f g ∈ I implies f ∈ I or

Definition 8.12

g m ∈ I for some m. Example 8.13

1. 2.

If P = f ⊂ k[x1 , . . . , xn ] is principal and prime then P M is primary. If m = x1 , . . . , xn ⊂ k[x1 , . . . , xn ] and I ⊂ k[x1 , . . . , xn ] is an ideal with m M ⊂ I for some M > 0 then I is primary. We prove the second assertion. Suppose that f g ∈ I and g m ∈ I for any m. It follows that g ∈ m and g(0, . . . , 0) = 0. We claim the multiplication μg : k[x1 , . . . , xn ]/m M → k[x1 , . . . , xn ]/m M h → hg

120


is injective: if h has Taylor expansion about the origin h = h d + higher-order terms,

h d = 0,

then gh has expansion gh = g(0, . . . , 0)h d + higher-order terms, with the first term nonzero. The quotient k[x1 , . . . , xn ]/m M is finite-dimensional, so μg is also surjective. Since μg (I ) ⊂ I , we must have μg (I ) ≡ I (mod m M ). In particular, f g ∈ I implies f ∈ I + m M = I . It is not generally true that if P ⊂ k[x1 , . . . , xn ] is prime then P m is primary! A counterexample is given in Example 8.18. Remark 8.14

Let Q be a primary ideal. Then the associated prime of Q.

Proposition 8.15

Proof

thus f ∈

√

√

Q is a prime ideal P, called

√ Let f g ∈ Q. Then f M g M ∈ Q and either f M ∈ Q or g Mm ∈ Q, and √ Q or g ∈ Q.

Proposition 8.16

Any irreducible ideal I in a Noetherian ring R is primary.

The converse does not hold – see Exercise 8.10 for a counterexample. Assume that I is irreducible and suppose that f g ∈ I . Consider the sequence of ideals

Proof

I ⊂ I : g ⊂ I : g 2 ⊂ . . . , which eventually terminates, so that I : g N = I : g N +1 for some N . We claim that (I + g N ) ∩ (I + f ) = I. The inclusion ⊃ is clear. Conversely, given an element h ∈ (I + g N ) ∩ (I + f ), we can write h = F1 + H1 g N = F2 + H2 f,

F1 , F2 ∈ I.

Since f g ∈ I we have g N +1 H1 ∈ I and thus g N H1 ∈ I and h ∈ I . Since I is irre ducible, the claim implies either f ∈ I or g N ∈ I .

8.3 PRIMARY IDEALS

121

Combining this with Proposition 8.3, we deduce the following result. Theorem 8.17 (Existence of primary decompositions)

For any ideal I in a Noethe-

rian ring R, we can write I = Q 1 ∩ Q 2 ∩ . . . ∩ Qr where each Q j is primary in R. This is called a primary decomposition of I . Example 8.18

AD − B 2 ,

Consider the ideal P ⊂ k[A, B, C, D, E, F] with generators AF − C 2 ,

D F − E 2,

AE − BC,

B E − C D,

B F − C E,

i.e., the two-by-two minors of the symmetric matrix ⎛

A M =⎝B C

B D E

⎞ C E ⎠. F

This ideal is prime: it consists of the equations vanishing on the image V of the map A3 (k) → A6 (k) (s, t, u) → (s 2 , st, su, t 2 , tu, u 2 ). We claim that P 2 is not primary. Consider the determinant det(M) = AD F − AE 2 − B 2 F + 2BC E − C 2 D, a cubic polynomial contained in P. However, det(M) ∈ P 2 because P 2 is generated by polynomials of degree 4. We have the following relations A det(M) = (AD − B 2 )(AF − C 2 ) − (AE − BC)2 B det(M) = (AD − B 2 )(B F − C E) − (AE − BC)(B E − C D) C det(M) = (B F − C E)(AE − BC) − (AF − C 2 )(B E − C D) D det(M) = (AD − B 2 )(D F − E 2 ) − (B E − C D)2 E det(M) = (AE − s BC)(D F − E 2 ) − (B F − C E)(B E − C D) F det(M) = (AF − C 2 )(D F − E 2 ) − (B F − C E)2 . If P 2 were primary then some power of each of the variables would be contained in P 2 , i.e., for some m A, B, C, D, E, Fm ⊂ P 2 . In particular, V = V (P 2 ) ⊂ {(0, 0, 0, 0, 0, 0)}, a contradiction.

122


8.4

Uniqueness of primary decomposition A primary decomposition I = Q 1 ∩ Q 2 ∩ . . . ∩ Qr , is weakly irredundant if none of the Q j is superfluous, i.e., Q j ⊃ Q 1 ∩ . . . Q j−1 ∩ Q j+1 . . . ∩ Q r for any j. For a weakly irredundant primary decomposition, the prime ideals P j = Q j are called the associated primes of I . Unfortunately, even weakly irredundant primary decompositions are not unique in general: Example 8.19

Consider two weakly irredundant primary decompositions I = x 2 , x y = x ∩ y, x 2 = Q 1 ∩ Q 2 = x ∩ y 2 , x 2 , x y = Q 1 ∩ Q 2 .

Here Q 1 = P1 is prime but Q 2 and Q 2 are not prime. They are primary with associated prime P2 =

Q2 =

Q 2 = x, y.

When the ideal is geometric in origin, e.g., I = I (V ) for some V ⊂ An (k), the primary components reflect the geometry: Example 8.20

Consider the weakly irredundant primary decomposition I = x z − y 2 , z − x y = x z − y 2 , z − x y, y − x 2 ∩ y, z = z − x y, y − x 2 ∩ y, z = Q 1 ∩ Q 2

Both Q 1 and Q 2 are prime: Q 1 consists of the equations vanishing on the image of the morphism φ

t → (t, t 2 , t 3 ) and Q 2 consists of the equations for the x-axis. The ideal I = I (V ) where V = image(φ) ∪ x-axis. See Remark 8.28 for a conceptual explanation of this example.

8.4 UNIQUENESS OF PRIMARY DECOMPOSITION

123

An ideal I ⊂ R is primary if and only if each zero divisor r ∈ R/I is nilpotent, i.e., r m = 0 for some m. When Q is primary, the nilpotents in R/Q √ are images of the associated prime P = Q.

Proposition 8.21

Proof

By definition, I is primary if and only if f g ∈ I ⇒ f ∈ I or g m ∈ I

for some m, i.e., f g = 0(mod I ) ⇒ f = 0(mod I )

or

g m = 0(mod I ),

which is the first assertion. Thus when Q is primary, all zero divisors in R/Q come √ from P = Q. Let R be Noetherian and I ⊂ R an ideal with weakly irredundant primary decomposition

Theorem 8.22 (Uniqueness Theorem I)

I = Q 1 ∩ Q 2 . . . ∩ Qr . The associated primes are precisely those primes which can be expressed as P = I : f for some f ∈ R. In particular, the associated primes are uniquely determined by I . Proof

From the definition of the quotient ideal we see that I : f = Q 1 : f ∩ . . . ∩ Q r : f .

Taking radicals and applying Exercise 7.18, we find that I : f = Q 1 : f ∩ . . . ∩ Q r : f . Furthermore, Proposition 8.21 gives

Qi : f =

Pi if f ∈ Q i R if f ∈ Q i .

By the irredundance assumption, there exists f ∈ Q j with f ∈ ∩i= j Q i . Then I : f = Q j : f = Pj , and each associated prime has the desired form. √ Conversely, suppose that P = I : f for some f ∈ R. Then we have P = ∩ j: f ∈ Q j P j .

124


To finish, we need the following fact: Let R be a ring and P ⊂ R a prime ideal which can be expressed as a finite intersection of primes

Lemma 8.23

P = ∩sj=1 P j . Then P = P j for some j. Suppose, on the contrary, that P P j for each j. Products are contained in intersections, so we have

Proof

P = ∩sj=1 P j ⊃ P1 . . . Ps . Pick g j ∈ P j , g j ∈ P so the product g = contradicts the fact that P is prime.

s j=1

g j ∈ P but each g j ∈ P. This

Let I ⊂ R be an ideal in a Noetherian ring. The zero divisors in R/I are the images of the associated primes for I .

Corollary 8.24

The proof above shows that every element of an associated prime is a zero divisor. Conversely, suppose we have ab = 0 in R/I with a, b = 0. Then for some j, a ∈ Q j but ab ∈ Q j . Then b is a zero divisor in Q j (or is contained in Q j ), and b ∈ P j by Proposition 8.21.

Proof

Let I ⊂ R be an ideal in a Noetherian ring. An associated prime is minimal if it does not contain any other associated prime; otherwise, it is embedded.

Definition 8.25

Let I ⊂ R be an ideal in a Noetherian ring. Then intersection of the minimal associated primes of I .

Proposition 8.26

Proof

√

I is the

Fix a primary decomposition I = Q 1 ∩ . . . ∩ Qr

√ with associated primes Pi = Q i , i = 1, . . . , r . Exercise 7.18 implies √ I = Q 1 ∩ . . . Q r = P1 ∩ . . . ∩ Pr . Of course, excluding the embedded primes does not change the intersection.

We retain the notation of Example 8.18. We showed that P 2 is not primary by computing

Example 8.27

det(M)m ⊂ P 2 ,

m := A, B, C, D, E, F,


125

but det(M) ∈ P 2 . It follows that

P 2 : det(M) = m

and Uniqueness Theorem I (8.22) implies m is an associated prime of P 2 . P is the unique minimal√associated prime of P 2 : Indeed, for any prime P and integer m > 0 we have P = P m (cf. Exercise 8.3.) By Proposition 8.26, the minimal associated primes of an ideal are the associated primes of its radical. The only associated primes of P 2 are m and P. The proof uses some Lie theory: Essentially, one classifies prime ideals in k[A, B, C, D, E, F] (the 3 × 3 symmetric matrices) invariant under similarities, i.e., the equivalence relation M ∼ T M T t where T is a 3 × 3 invertible matrix. This is carried out in [35, Prop. 4.15]; see also Exercise 8.12. Minimal associated primes have a nice interpretation for ideals arising from geometry. Let V ⊂ An (k) be a variety with irredundant decomposition into irreducible components

Remark 8.28

V = V1 ∪ . . . ∪ Vr ,

Vi ⊂ V j , i = j.

In Exercise 7.4 we saw that I (V ) ⊂ k[x1 , . . . , xn ] is radical. Proposition 3.6 gives the decomposition I (V ) = I (V1 ) ∩ . . . ∩ I (Vr ),

I (Vi ) ⊃ I (V j ), i = j,

and each I (V j ) is prime by Theorem 6.5. Thus the I (V j ), j = 1, . . . , r are the minimal associated primes of I (V ). We can sharpen this over algebraically closed fields: Let k be algebraically closed and I ⊂ k[x1 , . . . , xn ] an ideal. The minimal associated primes of I correspond to irreducible components of V (I ).

Corollary 8.29

√ Proposition 8.26 expresses I as the intersection of the minimal associ√ ated primes of I . The Hilbert Nullstellensatz gives I = I (V (I )), and we know that the minimal associated primes of I (V ) correspond to the components of V .

Proof

Based on this geometric intuition, we might expect the primary components associated with minimal associated primes to play a special rôle. We shall prove a strong uniqueness theorem for these components. Let Q 1 and Q 2 be primary ideals with √ Q 1 ∩ Q 2 is also primary with Q 1 ∩ Q 2 = P.

Proposition 8.30

√

Q1 =

√

Q 2 = P. Then

126


Let f g ∈ Q 1 ∩ Q 2 . Suppose that f ∈ Q 1 ∩ Q 2 , and assume that f ∈ Q 1 . √ Then g ∈ Q 1 for some m, and thus g ∈ P. Since Q 2 = P, P M ⊂ Q 2 for some large M and g M ∈ Q 2 . Consequently, g max(m,M) ∈ Q 1 ∩ Q 2 .

Proof

m

Suppose we have a weakly irredundant primary decomposition I = Q 1 ∩ Q 2 ∩ . . . ∩ Qr such that two primary components have the same associated prime, i.e. Pi = P j for some i = j. The proposition allows us to replace Q i and Q j with the single primary Qi ∩ Q j . Definition 8.31

A primary decomposition I = Q 1 ∩ Q 2 ∩ . . . ∩ Qr

is (strongly) irredundant if it is weakly irredundant and the associated primes Pi are distinct. Let I ⊂ R be an ideal in a Noetherian ring with irredundant primary decomposition

Theorem 8.32 (Uniqueness Theorem II)

I = Q 1 ∩ Q 2 ∩ . . . ∩ Qr . Suppose that P j is a minimal associated prime. Then there exists an element a ∈ ∩i= j Pi , a ∈ P j , and for each such element we have Q j = ∪m (I : a m ). In particular, the primary components associated to minimal primes are unique. If there exists no element a with the desired properties then P j ⊃ ∩i= j Pi . Repeating the argument for Lemma 8.23, we find that P j ⊃ Pi for some i = j, which contradicts the minimality of P j . Given such an element, for any r ∈ R there exists an m such that a m r ∈ Q i , i = j. On the other hand, a is not a zero-divisor modulo Q j , so if r ∈ Q j then a m r ∈ Q j for any m. Thus Proof

am r ∈ I = Q 1 ∩ Q 2 ∩ . . . ∩ Qr for m 0 if and only if r ∈ Q j .

The elements employed in Theorem 8.32 are easy to interpret when the ideal comes from geometry. As in Remark 8.28, suppose that I = I (V ) for a

Remark 8.33


127

variety V ⊂ An (k) with irreducible components V = V1 ∪ . . . ∪ Vr ,

Vi ⊂ V j , i = j.

The associated primes of I (V ) are P j = I (V j ), j = 1, . . . , r and ∩i= j Pi = {a ∈ k[V ] : a ≡ 0 on each Vi , i = j}. Theorem 8.32 requires an element a ∈ k[V ] vanishing on Vi for each i = j but not on V j . Theorem 8.32 is quite useful for computations. Once we have found an element satisfying its hypotheses, we can effectively compute the corresponding primary component. The key tool is the following fact: Proposition 8.34

Let R be a ring, I ⊂ R an ideal, and a ∈ R. Suppose that for

some integer M ≥ 0 I : a M = I : a M+1 , where by convention a 0 = 1. Then we have ∪m (I : a m ) = I : a M . Proof

It suffices to show that I : a M+1 = I : a M+2 .

Indeed, this will imply I : a M = I : a M+1 = I : a M+2 = I : a M+3 = . . . , which is our result. Pick r ∈ (I : a M+2 ), so that ra M+2 ∈ I . We therefore have ra ∈ I : a M+1 and our hypothesis guarantees then that ra ∈ I : a M . We deduce that (ra)a M = ra M+1 ∈ I . We retain the notation of Example 8.27. What is the distinguished primary ideal Q 1 ⊃ P 2 corresponding to the minimal prime P? We shall show that

Example 8.35

Q 1 = P 2 + det(M). First we establish that det(M) ∈ Q 1 ; if det(M) ∈ Q 1 then elements of m would be zero √ divisors (and hence nilpotents) modulo Q 1 , whence m ⊂ Q 1 = P, a contradiction. To apply Theorem 8.32, we need an element a ∈ m, a ∈ P; we take a = A.

128


Using the algorithms for computing quotient ideals and intersections (Algorithm 8.10 and Proposition 6.19), we compute P 2 : A = AE 2 − 2BC E − AF D + C 2 D + B 2 F, D 2 F 2 − 2D F E 2 + E 4 , B F 2 D − B F E 2 − C E D F + C E 3, B E D F − B E 3 − C D2 F + C D E 2, B 2 F 2 − 2B FC E + C 2 E 2 , B E 2 C − AE 3 − C D B F + AE D F, B 2 E 2 − AD E 2 − B 2 D F + AD 2 F, C 3 E − C 2 B F − AFC E + AF 2 B, B EC 2 − AE 2 C − B 2 C F + B E AF, C E B 2 − ADC E − B 3 F + AD B F, D B 2 C − AD 2 C − E B 3 + AD B E, C 4 − 2AFC 2 + A2 F 2 , BC 3 − AEC 2 − BC AF + A2 E F, A2 E 2 − 2AE BC + B 2 C 2 , B 3 C − AD BC − B 2 AE + A2 D E, A2 D 2 − 2AD B 2 + B 4 = det(M) + P 2 and P 2 : A2 = det(M) + P 2 . Proposition 8.34 implies that Q 1 = ∪m P 2 : Am = det(M) + P 2 . One last intersection computation implies that our primary decomposition is P 2 = (P 2 + det(M)) ∩ m4 . The component Q 2 = m4 is not unique: we could take Q 2 = Q 2 + g,

g ∈ m3 , g ∈ det(M);

any proper ideal Q 2 ⊃ mn is m-primary.

8.5

An application to rational maps Let ρ : V W denote a rational map of affine varieties. The indeterminacy ideal Iρ ⊂ k[V ] is defined

Definition 8.36

{r ∈ k[V ] : rρ ∗ f ∈ k[V ] for each f ∈ k[W ]}. We leave it to the reader to check that this is an ideal! This is compatible with our previous definition of the indeterminacy locus: Proposition 8.37

The indeterminacy locus of ρ is equal to V (Iρ ).

8.5 AN APPLICATION TO RATIONAL MAPS

129

Choose realizations V ⊂ An (k) and W ⊂ Am (k), so that ρ is determined by the pull-back homomorphism (see Corolllary 3.46)

Proof

ρ ∗ : k[y1 , . . . , ym ] → k(V ). Suppose first that v ∈ V (Iρ ). Let ρ : An (k) Am (k) be a rational map admissible along V such that ρ |V = ρ. Express each coordinate ρ j = f j /g, with f j , g ∈ k[x1 , . . . , xn ] and g not dividing zero in k[V ]. We have gρ ∗ k[W ] ⊂ k[V ] and thus g ∈ Iρ . Hence g(v) = 0 and ρ is not defined at v. In particular, v is contained in the indeterminacy locus of ρ. Assume that v ∈ V (Iρ ). We claim there exists an element g¯ ∈ Iρ such that g¯ ¯ ¯ ∗ y j ∈ k[V ] for each j. does not divide zero in k[V ] and g(v) = 0. It follows that gρ We can then choose polynomials g, f 1 , . . . , f m ∈ k[x1 , . . . , xn ] such that g ≡ g¯ and ¯ ∗ y j in k[V ]. The rational map f j ≡ gρ ρ : An (k) → Am (k) (x1 , . . . , xn ) → ( f 1 /g, . . . , f m /g) is admissible on V and induces ρ. To prove the claim, we will require the following Let R be a ring and P1 , . . . , Ps prime ideals in R. If I ⊂ R is an ideal with I ⊂ ∪s=1 P then I ⊂ P for some .

Lemma 8.38 (Prime avoidance)

We first give the application: The zero divisors of k[V ] are the union of the primes associated to the zero ideal in k[V ]; they correspond to functions vanishing on at least one irreducible component of V (see Corollary 8.24 and Remark 8.28.) Let R = k[V ], P1 , . . . , Ps−1 be the primes associated to zero, and Ps the maximal ideal corresponding to v. Not every element of Iρ is a zero divisor, as ρ has a representative by a rational map on affine space admissible along V . In particular, Iρ ⊂ ∪s−1 =1 P and Iρ ⊂ Ps = I (v). The lemma says that Iρ ⊂ ∪s=1 P , so we can pick g¯ ∈ Iρ such that g¯ neither divides zero nor vanishes at v. The argument is by induction: if I ⊂ ∪s=1 Pl we show that I is contained in the union of some collection of s − 1 primes. Suppose, on the contrary, that for each = 1, . . . , s there exists r ∈ I with r ∈ ∪ j= P j . We must then have r ∈ P for each . Consider the element Proof of lemma:

t = r1 + r2 . . . rs ∈ I ⊂ P1 ∪ P2 ∪ . . . ∪ Ps . If t ∈ P1 then r2 . . . rs ∈ P1 and r ∈ P1 for some = 1. On the other hand, suppose t ∈ P j for some j > 1; since r2 . . . rs ∈ P j , we deduce r1 ∈ P j . In either case, we reach a contradiction.

130


Retain the notation of Proposition 8.37. The following are

Proposition 8.39

equivalent: 1. 2. 3.

ρ is a morphism; ρ ∗ k[W ] ⊂ k[V ]; Iρ = k[V ]. The equivalence of the first two conditions follows from Corollaries 3.32 and 3.46: rational maps V W (resp. morphisms V → W ) correspond to homomorphisms k[W ] → k(V ) (resp. k[W ] → k[V ]). For the third, Iρ = k[V ] precisely when 1 ∈ Iρ , i.e., when ρ ∗ f ∈ k[V ] for each f ∈ k[W ].

Proof

Our next result sharpens Proposition 8.37 and puts Example 3.49 in a general framework: Let k be algebraically closed and consider a rational map of affine varieties ρ : V W . Then the following are equivalent:

Proposition 8.40

1. 2.

ρ is a morphism; the indeterminacy locus of ρ is empty. The first condition obviously defines the second; we prove the converse. Let Iρ ⊂ k[V ] denote the indeterminacy ideal; Proposition 8.37 implies V (Iρ ) = ∅. Realize V ⊂ An (k) as a closed subset with quotient homomorphism q : k[x1 , . . . , xn ] k[V ], and set I = q ∗ Iρ so that V (I ) = V (Iρ ). Since V (I ) = ∅, the Nullstellensatz (Theorem 7.5) implies I = k[x1 , . . . , xn ] and thus Iρ = k[V ]. It follows that ρ is a morphism.

Proof

Given a rational map ρ : V Am (k), we seek an algorithm for computing the indeterminacy ideal Iρ = {r ∈ k[V ] : rρ ∗ k[Am (k)] ⊂ k[V ]} ⊂ k[V ], or, more precisely, its preimage in k[x1 , . . . , xn ] JV,ρ := {h ∈ k[x1 , . . . , xn ] : h (mod I (V )) ∈ Iρ }. We have Iρ = {r ∈ k[V ] : r f j /g j ∈ k[V ], j = 1, . . . , m} = ∩M j=1 {r ∈ k[V ] : r f j /g j ∈ k[V ]} = ∩M j=1 {r ∈ k[V ] : r f j ∈ g j k[V ]} = ∩M j=1 g j : f j ,

8.6 EXERCISES

131

which implies JV,ρ = ∩mj=1 (g j + I (V )) : ( f j + I (V )).

(8.1)

The last step is an application of the following general fact Let ψ : R S be a surjective ring homomorphism, J1 , J2 ⊂ S ideals with preimages Ii = ψ −1 (Ji ) ⊂ R. Then

Lemma 8.41

I1 : I2 = ψ −1 (J1 : J2 ). The surjectivity of ψ implies ψ(Ii ) = Ji for i = 1, 2. We prove I1 : I2 ⊂ ψ −1 (J1 : J2 ) first. Given r ∈ R with r I2 ⊂ I1 , applying ψ yields ψ(r )ψ(I2 ) ⊂ ψ(I1 ). It follows then that ψ(r ) ∈ J1 : J2 . We turn to the reverse implication. Take r ∈ R with ψ(r ) ∈ J1 : J2 . For each w ∈ I2 we have ψ(r )ψ(w) = ψ(r w) ∈ J1 , hence r w ∈ ψ −1 (J1 ) = I1 . It follows that r ∈ I1 : I2 .

Proof

Formula 8.1 allows us to compute JV,ρ by iterating our previous algorithms for computing quotients and intersections (Algorithm 8.10 and Proposition 6.19.)

8.6

Exercises

8.1

Flesh out the details in Example 8.6. Prove x 2 , x y, y 2 = y, x 2 ∩ y 2 , x = y + x, x 2 ∩ x, (y + x)2

8.2

using our algorithm for computing intersections. Verify that the ideals appearing are irreducible. Let C = {(x1 , x2 ) : x13 − x22 = 0} ⊂ A2 (Q) and consider the birational parametrization φ : A1 (Q) → C t → (t 2 , t 3 ).

8.3

8.4 8.5

Use the method of §8.5 to compute the indeterminacy ideal Iφ −1 ⊂ Q[C]. √ Let P ⊂ R be a prime ideal and m a positive integer. Show that P m = P. In particular, if R is Noetherian then P is a minimal associated prime of P m . In this case, the primary component of P m associated to P is called the mth symbolic power of P. Let R be Noetherian and m ⊂ R a maximal ideal. Let Q ⊂ R be an ideal such that m M ⊂ Q ⊂ m for some M. Show that Q is m-primary. (a) Show that the ideal Q = x y − z 2 , x ⊂ C[x, y, z] is primary. Describe the associated prime.

132


(b) Compute an irredundant primary decomposition and associated primes of the ideal I = z 2 , yz, x z, y 2 − x 2 (x + 1) ⊂ Q[x, y, z]. (c) Consider the ideal I = x y, x z ⊂ Q[x, y, z].

8.6

Describe the irreducible components of V (I ) and compute the primary decomposition of I . Consider the ideals P1 = x, y,

8.7

P2 = x, z,

Show that the Pi are prime and compute an irredundant primary decomposition and the associated primes of the product P1 P2 P3 . Consider the ideals I1 = x, y,

8.8 8.9

P3 = y, z ⊂ Q[x, y, z].

I2 = y, z.

(a) Compute the intersection I1 ∩ I2 , using the Gröbner basis algorithm. (b) Find a primary decomposition for I1 I2 . Does it have an embedded prime? (c) Does I1 I2 = I1 ∩ I2 ? Let I1 , I2 ⊂ k[x1 , . . . , xn ] be ideals. Show that I1 : I2 = I1 if I2 is not contained in any of the associated primes of I1 . Show that a principal ideal in a polynomial ring has no embedded associated primes. On the other hand, if R = k[x1 , x2 , x3 , x4 ]/ x1 x4 − x2 x3 , x23 − x12 x3 , x33 − x2 x42 show that the principal ideal x2 ⊂ R

8.10 8.11

has embedded associated primes. Give an example of a primary ideal Q ⊂ k[x1 , . . . , xn ] which is not irreducible, i.e., Q = I1 ∩ I2 , I1 , I2 = Q. (a) Find generators for the ideals J (X i ) for the following closed subsets: X 1 = {x0 = x1 = 0} ∪ {x1 = x2 = 0} ∪ {x2 = x3 = 0} ∪ {x3 = x0 = 0} ⊂ P3 X 2 = ({x0 = x1 = 0} ∪ {x2 = x3 = 0}) ∩ {x0 + x1 + x2 + x3 = 0} ⊂ P3 (b) Is the ideal J (X 2 ) equal to J = (x0 , x1 ∩ x2 , x3 ) + x0 + x1 + x2 + x3 ?

8.6 EXERCISES

8.12

133

√ (c) Compute the primary decompositions of J (X 2 ) and J . Hint: What is J : f for f = x0 + x1 − x2 − x3 ? (d) How do you interpret the embedded prime of J ? Two symmetric n × n matrices M, M are similar if there exists an invertible n × n matrix T with M = T M T t . (a) Show that similar matrices have the same rank. (b) Show that complex matrices of the same rank are similar. Identify An(n+1)/2 (C) with the n × n complex symmetric matrices M = (m i j ). An ideal I ⊂ C[m i j ] is invariant under similarity if, for every invertible matrix T , f (M) ∈ I ⇒ f (T M T t ) ∈ I. (c) Let Sr ⊂ An(n+1)/2 (C) denote the matrices of rank r . Show that I (Sr ) is prime and invariant. Hint: Consider the morphism 2

An → An(n+1)/2 T → T M T t with image Sr with r = rank(M). (d) Show that each invariant prime ideal P ⊂ C[m i j ] equals I (Sr ) for some r .

9 Projective geometry

Projective geometry arose historically out of plane geometry. It is very fruitful to introduce points ‘at infinity’ where parallel lines intersect. This leads to a very elegant approach to incidence questions, where points and lines are on an equal and symmetric footing. In the context of algebraic geometry, points at infinity are crucial in the statement of uniform results, like Bezout’s Theorem on the intersection of two plane curves. However, to do projective geometry we must leave the realm of affine varieties. A projective variety is constructed by gluing a number of affine varieties together. There are many subtle issues that arise, especially when the base field is not algebraically closed. These are deferred to the end of this chapter. Thankfully, there is an extremely concrete approach to projective geometry using the algebra of homogeneous polynomials. This allows us to apply many of the computational techniques developed for affine varieties to projective varieties, with minor modifications. Indeed, concrete problems in affine geometry often become more transparent once they are translated into projective language.

9.1

Introduction to projective space Projective n-space Pn (k) is the set of all lines in affine space containing the origin 0 ∈ ⊂ An+1 (k). Each such line takes the form span(a0 , . . . , an ) = λ(a0 , . . . , an ),

λ ∈ k,

where (a0 , . . . , an ) ∈ An+1 (k) − {0}. Two elements (a0 , . . . , an ), (a0 , . . . , an ) ∈ An+1 (k) span the same line if (a0 , . . . , an ) = λ(a0 , . . . , an ), λ ∈ k ∗ . Thus we can identify projective space with equivalence classes Pn (k) = (An+1 (k) − {0})/ ∼ 134

9.1 INTRODUCTION TO PROJECTIVE SPACE

135

where (a0 , . . . , an ) ∼ (a0 , . . . , an )

if (a0 , . . . , an ) = λ(a0 , . . . , an ),

λ ∈ k∗.

We’ll use the notation [a0 , . . . , an ] to denote one of the resulting equivalence classes. There is a natural way to parametrize ‘most’ lines in An+1 (k) by an affine space of dimension n. For each i = 0, . . . , n, consider the lines of the form Ui = {[a0 , . . . , an ] : ai = 0} ⊂ Pn (k). The vanishing (or nonvanishing) of ai is compatible with the following equivalence relation: when [a0 , . . . , an ] ∼ [a0 , . . . , an ] ai = 0 if and only if ai = 0. We have a function ψi : Ui → An (k) [a0 , . . . , an ] → (a0 /ai , . . . , ai−1 /ai , ai+1 /ai , . . . , an /ai ) with inverse ψi−1 : An (k) → Ui

(b0 , . . . , bi−1 , bi+1 , . . . , bn ) → [b0 , . . . , bi−1 , 1 , bi+1 , . . . , bn ] . ith place

The function ψi identifies lines in Ui with points in the affine space An (k) with coordinates b0 , . . . , bi−1 , bi+1 , . . . , bn . For any point in Pn (k) some ai = 0, so we can express Pn (k) = U0 ∪ U1 ∪ . . . ∪ Un , Example 9.1

Ui An (k).

(9.1)

The line span(a0 , a1 ) ⊂ A2 (k) is defined by the linear equation a1 x0 − a0 x1 = 0.

When a0 = 0 we can divide through by a0 to get x1 = b1 x0 ,

b1 = a1 /a0 ;

these are the lines in U0 . When a1 = 0 we get x 0 = b0 x 1 ,

b0 = a0 /a1 ,

corresponding to the lines in U1 . What are the lines ‘left out’ by our distinguished subsets Ui ⊂ Pn (k)? The complement Hi := Pn (k)\Ui = {[a0 , . . . , ai−1 , 0, ai+1 , . . . , an ] : (a0 , . . . , ai−1 , ai+1 , . . . , an ) = 0}

136

PROJECTIVE GEOMETRY x0 = 1 1

U1= A

x1 = 1

(0, 0)

1

U0 = A

Figure 9.1

Two A1 s parametrizing lines in A2 .

consists of the lines through the origin in the affine subspace 0 ∈ ⊂ An (k) = {ai = 0} ⊂ An+1 (k). We therefore may express Pn (k) = Ui ∪ Hi = An (k) ∪ Pn−1 (k), where we interpret P0 (k) as a point. How do the identifications ψi : Ui An (k) fit together? Restrict to the intersections ψj

An (k)

Ui ∩ U j ψi

An (k)

with i < j for notational simplicity. We consider the compositions ρi j = ψi ◦ ψ −1 j ψ j (Ui ∩ U j ) ||

ψi ◦ψ −1 j

→ ρi j

ψi (Ui ∩ U j ) ||

An (k) \ {bi = 0} → An (k) \ {bj = 0}. −1 −1 Since (ψi ◦ ψ −1 j ) ◦ (ψ j ◦ ψ ) = ψi ◦ ψ we have the compatibility condition

ρi j ◦ ρ j = ρi

9.2 HOMOGENIZATION AND DEHOMOGENIZATION

137

for all i, j, . We explicitly compute ρi j ψ −1 j

(b0 , . . . , b j−1 , b j+1 , . . . , bn ) −→ [b0 , . . . , b j−1 , 1, b j+1 , . . . , bn ] = [b0 /bi , . . . , bi−1 /bi , 1, bi+1 /bi , . . . , b j−1 /bi , 1/bi , b j+1 /bi , . . . , bn /bi ] ψi

−→ (b0 /bi , . . . , bi−1 /bi , bi+1 /bi , . . . , b j−1 /bi , 1/bi , b j+1 /bi , . . . , bn /bi ) which defines a birational map An (k) An (k). Eliminating indeterminacy using Proposition 3.47, we get an isomorphism of affine varieties ρi j

An (k) \ {bi = 0} → An (k) \ {bj = 0} || || ∼ n n A (k)bi → A (k)bj . For P1 (k) we get a single birational map

Example 9.2

ρ01 : A1 (k) A1 b0 → 1/b0 . For P2 (k) we get ρ01 : A2 (k) A2 (k) (b0 , b2 ) → b0−1 , b2 /b0 , ρ02 : A2 (k) A2 (k) (b0 , b1 ) → b1 /b0 , 1/b0 , ρ12 : A2 (k) A2 (k) (b0 , b1 ) → b0 /b1 , 1/b1 satisfying ρ02 = ρ01 ◦ ρ12 . Let V be a finite-dimensional k-vector space of dimension n + 1. The projectivization of V is defined

Remark 9.3 (Coordinate-free approach)

P(V ) = {one-dimensional vector subspaces ⊂ V }. While this is just P n (k), for some constructions it is useful to keep track of the underlying vector-space structure. For example, P(k[x0 , . . . , xn ]d ) denotes the projective space modeled on the homogeneous forms of degree d.

9.2

Homogenization and dehomogenization Each polynomial f ∈ k[x0 , . . . , xn ] can be decomposed into homogeneous pieces f = F0 + F1 + · · · + Fd ,

d = deg( f ),

138

PROJECTIVE GEOMETRY

i.e., each F j is homogeneous of degree j in x0 , . . . , xn . An ideal J ⊂ k[x0 , . . . , xn ] is homogeneous if it admits a collection of homogeneous generators. Equivalently, if a polynomial is in a homogenous ideal then each of its homogeneous pieces is in that ideal (see Exercise 9.1). Dehomogenization with respect to xi is defined as the homomorphism μi : k[x0 , . . . , xn ] → k[y0 , . . . , yi−1 , yi+1 , . . . , yn ] xi → 1 x j → y j , j = 1. For f ∈ k[y0 , . . . , yi−1 , yi , . . . , yn ], the preimage μi−1 ( f ) contains D xi f (x0 /xi , . . . , xi−1 /xi , xi+1 /xi , . . . , xn /xi ) : D ≥ deg( f ) and equals the affine span of these polynomials. The homogenization of f with respect to xi is defined deg( f )

F(x0 , . . . , xn ) := xi

f (x0 /xi , . . . , xi−1 /xi , xi+1 /xi , . . . , xn /xi ).

The homogenization of an ideal I ⊂ k[y0 , . . . , yi−1 , yi+1 , . . . , yn ] is the ideal generated by the homogenizations of each f ∈ I . Given an ideal I = f 1 , . . . , fr , the homogenization J need not be generated by the homogenizations of the elements, i.e.,

deg( f j ) J = xi f j (x0 /xi , . . . , xi−1 /xi , xi+1 /xi , . . . , xn /xi ) j=1,...,r in general. Example 9.4

Consider

I = y2 − y12 , y3 − y1 y2 = f 1 , f 2

and dehomogenize with respect to x0 μ0 : k[x0 , x1 , x2 , x3 ] → k[y1 , y2 , y3 ]. The homogenization of the elements f 1 , f 2 gives an ideal

x2 x0 − x12 , x3 x0 − x1 x2 J. The polynomial h = x22 − x1 x3 ∈ J because y22 − y1 y3 = y2 f 1 − y1 f 2 , but h is not contained in the ideal generated by the homogenizations of f 1 and f 2 . A monomial order on k[y1 , . . . , yn ] is graded if it is compatible with the partial order induced by degree, i.e., y α > y β whenever |α| > |β|.

Definition 9.5

For instance, lexicographic order is not graded, because small degree monomials in the first variable precede large degree monomials in the subsequent variables.

9.2 HOMOGENIZATION AND DEHOMOGENIZATION

139

Let I ⊂ k[y1 , . . . , yn ] be an ideal and J ⊂ k[x0 , . . . , xn ] its homogenization with respect to x0 . Suppose that f 1 , . . . , fr is a Gröbner basis for I with respect to some graded order >. Then the homogenizations F1 , . . . , Fr of f 1 , . . . , fr generate J . Theorem 9.6

We first introduce an order >x on the x-monomials derived from >. We

Proof

define x0α0 x1α1

. . . xnαn

β β >x x 0 0 x 1 1

. . . xnβn

⇔

β

β

if y1α1 . . . ynαn > y1 1 . . . yn n , β

β

or y1α1 . . . ynαn = y1 1 . . . yn n and α0 > β0 .

We leave it as an exercise to verify that this defines a monomial order. The theorem will follow once we show that F1 , . . . , Fr form a Gröbner basis for J with respect to >x ; Corollary 2.14 guarantees they generate J . Let G ∈ k[x0 , . . . , xn ] be homogeneous with dehomogenization g = deg(G)−deg(g) α1 μ0 (G). If LT> (g) = cy1α1 . . . ynαn then LT>x (G) = cx0 x1 . . . xnαn ; in particular, LT> (g) = μ0 (LT>x (G)).

Lemma 9.7

Let G denote the homogenenization of g with respect to x0 , so that G = deg(G)−deg(g) x0 G . It suffices to show that LT>x (G ) = cx1α1 . . . xnαn . Since > is graded, LT> (g) = cy1α1 . . . ynαn has degree equal to deg(g). Thus cx1α1 . . . xnαn is a term of G . Consider terms in G in which only x1 , x2 , . . . , xn appear; the leading term of G is one of these. Indeed, terms containing x0 dehomogenize to terms in y1 , . . . , yn of degree < deg(g), and thus are smaller than monomials of degree deg(g) in x1 , . . . , xn . The order induced on monomials in y1 , . . . , yn by > coincides with the order induced on monomials in x1 , . . . , xn by >x , so the leading terms of g and G coincide. This completes the proof of the lemma. μ0 (J ) ⊂ I .

Lemma 9.8

Choose a homogeneous H ∈ J and express

deg g H= A j g j (x1 /x0 , . . . , xn /x0 )x0 j ,

g j ∈ I, A j ∈ k[x0 , . . . , xn ].

j

Dehomogenizing with respect to x0 , we obtain

μ0 (H ) = A j (1, y1 , . . . , yn )g j , j

so μ0 (H ) ∈ I and the lemma is proven. Suppose that H is a homogeneous polynomial in J . It suffices to prove that LT>x (H ) is divisible by LT>x (F j ) for some j. By the second lemma above h = μ0 (H ) ∈ I . Since f 1 , . . . , fr are a Gröbner basis for I we have LT> ( f j )|LT> (h) for some j. Applying the first lemma twice, we conclude LT>x (F j )|LT>x (H ).

140

PROJECTIVE GEOMETRY

9.3

Projective varieties A projective variety X ⊂ Pn (k) is a subset such that, for each distinguished Ui An (k), i = 0, . . . , n, the intersection Ui ∩ X ⊂ Ui is affine.

Definition 9.9

X ⊂ Pn (k) is Zariski closed if X ∩ Ui is closed in each distinguished Ui . For any subset S ⊂ Pn (k), the projective closure S ⊂ Pn (k) is defined as the smallest closed subset containing S.

Definition 9.10

A projective variety X ⊂ Pn (k) is reducible if it can be expressed as a union of two closed proper subsets

Definition 9.11

X = X 1 ∪ X 2,

X 1 , X 2 X.

It is irreducible if there is no such representation. We describe a natural way to get large numbers of projective varieties: Let F ∈ k[x0 , . . . , xn ] be homogeneous of degree d. Then there is a projective variety

Proposition 9.12

X (F) := {[a0 , . . . , an ] : F(a0 , . . . , an ) = 0} ⊂ Pn (k), called the hypersurface defined by F. More generally, given a homogeneous ideal J ⊂ k[x0 , . . . , xn ], we define X (J ) := {[a0 , . . . , an ] : F(a0 , . . . , an ) = 0 for each homogeneous F ∈ J }, the projective variety defined by J . Proof

(a0 , . . . , an )

Note that F does not yield a well-defined function on Pn (k): If = λ(a0 , . . . , an ) then F(a0 , . . . , an ) = λd F(a0 , . . . , an ).

However, we can make sense of the locus X (F) where F vanishes, because F(a0 , . . . , an ) = 0 if and only if F(a0 , . . . , an ) = 0. We check this is closed. On Ui we have xi = 0, so F = 0 if and only if xi−d F = 0. However, f := xi−d F = F(x0 /xi , . . . , xi−1 /xi , 1, xi+1 /xi , . . . , xn /xi ), is a well-defined polynomial on An (k). Hence Ui ∩ X (F) = V ( f ) ⊂ Ui An (k)

9.4 EQUATIONS FOR PROJECTIVE VARIETIES

141

is affine for each i. The final assertion is obtained by intersecting the X (F) for homogeneous F ∈ J . Example 9.13

1.

The 2 × 2 minors of ⎛

A ⎝B C

2.

B D E

⎞ C E⎠ F

define a closed subset of P5 (k) (which we’ll show is isomorphic to P2 (k)). The 2 × 2 minors of

A B

B C

C D

define a closed subset of P3 (k) (which we’ll show is isomorphic to P1 (k)). Given S ⊂ Pn (k), the homogeneous ideal vanishing along S is defined J (S) = F ∈ k[x0 , . . . , xn ] homogeneous : F(s) = 0 for each s ∈ S.

9.4

Equations for projective varieties Our goal is to prove that Proposition 9.12 is robust enough to produce every projective variety: Let X ⊂ Pn (k) be a projective variety. Then there exists a homogeneous ideal J such that X = X (J ).

Theorem 9.14

Our argument will yield an effective algorithm for computing J from the ideals I (X ∩ Ui ). Proposition 9.15

If S ⊂ Pn (k) then X (J (S)) = S.

It is clear that S ⊂ X (J (S)), so we have S ⊂ X (J (S)). We prove the reverse inclusion. Suppose that p ∈ S. There exists a distinguished open subset Ui ⊂ Pn (k) such that p ∈ Ui and xi ( p) = 0. Since Ui ∩ S is closed, there exists a polynomial f ∈ I (Ui ∩ S) that does not vanish at p. Let F be the homogenization of f ; we still have F( p) = 0. Note that F vanishes at all the points of S ∩ Ui and xi vanishes at each point of S not contained in Ui . Thus xi F ∈ J (S) and (xi F)( p) = 0, so p ∈ X (J (S)).

Proof

142

PROJECTIVE GEOMETRY

Let V ⊂ An (k) U0 ⊂ Pn be an affine variety with ideal I (V ) ⊂ k[y1 , . . . , yn ]. Let J ⊂ k[x0 , . . . , xn ] denote the homogenization of I (V ). Then J (V ) = J and X (J ) = V .

Proposition 9.16

Once we prove J (V ) = J , Proposition 9.15 implies X (J ) = V . We prove J (V ) ⊃ J . For each homogeneous G ∈ J , g := μ0 (G) ∈ I (V ) by Lemma 9.8. Thus G vanishes on [1, a1 , . . . , an ] whenever (a1 , . . . , an ) ∈ V , i.e., G ∈ J (V ). Conversely, suppose we are given a homogeneous H ∈ J (V ), so that H (1, b1 , . . . , bn ) = 0 for each (b1 , . . . , bn ) ∈ V . Writing h = μ0 (H ), we find that h ∈ I (V ). We can write

Proof

deg(H )

H = x0

deg(H )−deg(h)

h(x1 /x0 , . . . , xn /x0 ) = x0

H

where H is the homogenization of h. In particular, H is contained in the homogenization of I (V ). Remark 9.17

In this situation, the hyperplane H0 := {x0 = 0} = Pn (k) − U0

is often called the hyperplane at infinity. We have H0 ⊃ (V \ V ). Note that, in Proposition 9.16, we used the ideal of all functions vanishing on V rather than an arbitrary ideal vanishing on V . This is actually necessary; using any smaller ideal can lead to unwanted components at infinity. For instance, consider the ideal

Example 9.18

I = y12 + y24 ⊂ R[y1 , y2 ] which defines the origin in R2 . The homogenization is

J = x02 x12 + x24 , which defines a variety X (J ) = {x1 = x2 = 0} ∪ {x0 = x2 = 0} ⊂ P2 (R). This is strictly larger than the closure V (I ); it contains an extra point at infinity. The reason for this pathology is that we are working over a nonclosed field. Over the complex numbers, {x02 x12 + x24 = 0} ⊂ P2 (C) is a plane curve C with two singularities, [0, 1, 0] and [1, 0, 0]. These singularties are the only real points of C. The following result reduces the computation of the equations of a projective variety to the computation of the equations for affine varieties:

9.4 EQUATIONS FOR PROJECTIVE VARIETIES

143

Let X ⊂ Pn be closed, Ii = I (Ui ∩ X ) where Ui ⊂ Pn (k) is one of the distinguished open subsets, and Ji ⊂ k[x0 , . . . , xn ] the homogenization of Ii . Then we have

Proposition 9.19

J (X ) = J0 ∩ J1 ∩ . . . ∩ Jn . For each i = 0, . . . , n we have

Proof

X ⊃ Ui ∩ X and thus J (X ) ⊂ J (Ui ∩ X ) = Ji , n where the last equality follows from Proposition 9.16. In particular, J (X ) ⊂ ∩i=0 Ji . n Conversely, each x ∈ X is contained in some distinguished Ui , so X ⊂ ∪i=0 (X ∩ Ui ) and n J (X ) ⊃ ∩i=0 Ji .

Example 9.20

Consider

X = x02 + x12 = x22 ∪ {x0 = 0} ∪ {x1 = 0} ∪ {x2 = 0} ⊂ P2 (C). We have U0 ∩ X = 1 + y12 = y22 ∪ {y1 = 0} ∪ {y2 = 0} and hence

I (U0 ∩ X ) = y1 y2 1 + y12 − y22 with homogenization

J0 = x1 x2 x02 + x12 − x22 . Similarly,

J1 = x0 x2 x02 + x12 − x22 ,

J2 = x0 x1 x02 + x12 − x22

and therefore

J (X ) = J0 ∩ J1 ∩ J2 = x0 x1 x2 x02 + x12 − x22 . Example 9.21

Consider

X = (x0 , x1 , x2 ) : x02 = x0 x1 = x0 x2 = x12 = x1 x2 = x22 = 0 ⊂ P2 (k)

144

PROJECTIVE GEOMETRY

so that I (U0 ∩ X ) = 1,

I (U1 ∩ X ) = 1,

I (U2 ∩ X ) = 1

and J0 = J1 = J2 = J = 1. In particular, X is empty! This shows that we sometimes get additional equations by looking carefully in each of the distinguished open neighborhoods Ui .

9.5

Projective Nullstellensatz In the previous section, we gave an example of a homogeneous ideal J k[x0 , . . . , xn ] where X (J ) = ∅ and J (X (J )) = k[x0 , . . . , xn ]. Now we describe systematically when this happens. A homogeneous ideal J ⊂ k[x0 , . . . , xn ] is irrelevant if J ⊃ x0N , . . . , xnN for some N .

Definition 9.22

Fix m = x0 , . . . , xn . A homogeneous ideal J is irrelevant if and only if either of the following two equivalent conditions holds

Proposition 9.23

1. 2.

J ⊃ m M for some M; J is m-primary. The analysis of m-primary ideals in Example 8.13 shows the equivalence of the two conditions. It is clear that if the first condition holds then J is irrelevant. Conversly, if x0N , x1N , . . . , xnN ∈ J then every monomial in x0 , . . . , xn of degree at least (n + 1)N is contained in J .

Proof

Proposition 9.24 Proof

If J ⊂ k[x0 , . . . , xn ] is irrelevant then X (J ) = ∅.

By Proposition 9.23, we know that x0N , . . . , xnN ∈ J for some large N ,

hence X (J ) ⊂ x0N = x1N = . . . = xnN = 0 = {x0 = x1 = . . . = xn = 0} = ∅.

Let k be algebrically closed and J ⊂ k[x0 , . . . , xn ] be a homogeneous ideal. Then X (J ) = ∅ if and only if J is irrelevant.

Theorem 9.25 (Projective Nullstellensatz)

The implication ⇐ is Proposition 9.24. To prove ⇒, assume that X (J ) = ∅. Let Ii be the dehomogenization of J with respect to xi , i.e., Ii = μi (J ). For any g ∈ Ii with homogenization G, we have

Proof

deg(g)

xie xi

g(x0 /xi , . . . , xn /xi ) = xie G ∈ J

9.6 MORPHISMS OF PROJECTIVE VARIETIES

145

for some e > 0. Indeed, g is the dehomogenization of some homogeneous H ∈ J and we have H = xie G. We deduce Ui ∩ X = V (Ii ) because on Ui ∩ X , g vanishes exactly where G vanishes. Since Ui ∩ X = ∅, the affine Nullstellensatz II (Theorem 7.5) implies Ii = 1, and it follows that xie ∈ J . We conclude that J is irrelevant. Let k be algebraically closed and J ⊂ k[x0 , . . . , xn ] be homogeneous with variety X = X (J ). If H ∈ k[x0 , . . . , xn ] is homogeneous and vanishes on X then H N ∈ J for some N .

Theorem 9.26 (Projective Hilbert Nullstellensatz)

Retain the notation introduced in the proof of the last theorem. If h i = μi (H ) is the dehomogenization of H with respect to xi then h iNi ∈ Ii for some Ni by the affine Hilbert Nullstellensatz (Theorem 7.3). Rehomogenizing, we get xiei H Ni ∈ J for suitable ei > 0. Each monomial of sufficiently large degree is contained in x0e0 , . . . , xnen , so we have H M ∈ x0e0 , . . . , xnen for some M 0. It follows that

Proof

H M H max{N0 ,...,Nn } ∈ J.

9.6

Morphisms of projective varieties Recall that a morphism of affine spaces is a polynomial map φ : An (k) → Am (k) (x1 , . . . , xn ) → (φ1 , . . . , φm ),

φ j ∈ k[x1 , . . . , xn ].

Naively, we might try to define a morphism φ : P n (k) → P m (k) analogously φ : P n (k) → P m (k) [x0 , . . . , xn ] → [φ0 , . . . , φm ],

φ j ∈ k[x0 , . . . , xn ].

This only makes sense when the φ j are all homogeneous of degree d, in which case φ[λx0 , . . . , λxn ] = [λd φ0 , . . . , λd φm ] = [φ0 , . . . , φm ] = φ[x0 , . . . , xn ] and φ is well-defined on equivalence classes. Definition 9.27

A polynomial (rational) map φ : P n (k) P m (k)

is given by a rule [x0 , . . . , xn ] → [φ0 , . . . , φm ], where the φ j ∈ k[x0 , . . . , xn ] are all homogeneous of degree d ≥ 0.

146

PROJECTIVE GEOMETRY

Of course, φ is defined everywhere on P n (k) provided the φ j have only trivial common zeros, i.e., if φ0 (a0 , . . . , an ) = . . . = φm (a0 , . . . , an ) = 0

⇒

a0 = . . . = an = 0.

(†)

In other words, φ(a0 , . . . , an ) determines a point in P m (k) for each [a0 , . . . , an ] ∈ P n (k). The following example suggests why we might want to ask for a bit more: Consider the polynomial map

Example 9.28

φ : P2 (R) P1 (R) [x0 , x1 , x2 ] → x02 + x12 , x12 + x22 , which satisfies both of our naive conditions. However, the same rule does not define a function φ : P2 (C) → P1 (C). Over C, φ0 and φ1 have a common solution. Thus condition (†) does not behave well as the coefficient field is varied. We faced the same issue for the indeterminacy of rational maps over nonclosed fields (see Example 3.49 and Proposition 8.40.) We can avoid this problem by insisting that φ0 , . . . , φm be irrelevant. Proposition 9.24 then guarantees that φ is well-defined over any extension: Proposition 9.29

A polynomial map φ : P n (k) P m (k)

defines a morphism provided the ideal φ0 , . . . , φm is irrelevant. The irrelevance is not a necessary condition for the polynomial map to induce a morphism (see Exercise 9.10). This generalizes readily to arbitrary projective varieties X ⊂ P n (k), with a slight twist. For a polynomial map φ to specify a morphism φ : X → P m (k) it is not necessary that the φ j have no common solution anywhere on P n (k). We don’t care whether φ makes sense on the complement P n (k) \ X : Let X ⊂ P m (k) be a projective variety. A polynomial map φ : P (k) P (k) restricts to a morphism X → P m (k) provided J (X ) + φ0 , . . . , φm is irrelevant.

Proposition 9.30 n

m


147

This situation is in marked contrast to the affine case discussed in Chapter 3: given affine V ⊂ An (k) and W ⊂ Am (k), a morphism φ : V → W always extends to a morphism φ : An (k) → Am (k).

Remark 9.31

The proofs of Propositions 9.29 and 9.30 make reference to the general definitions of abstract varieties and morphisms in §9.8. Readers who wish to avoid this machinery might take Proposition 9.30 as a working definition of a morphism from a projective variety to projective space. 9.6.1 Projection and linear maps

We generalize the distinguished open affine subsets Ui introduced in the definition of projective space: n ci xi ∈ k[x0 , . . . , xn ] be a nonzero homogeneous Let L = i=0 linear form. Then the open subset

Proposition 9.32

U L = {[a0 , . . . , an ] : L(a0 , . . . , an ) = 0} ⊂ P n (k) is naturally an affine variety, isomorphic to An (k). This expanded inventory of affine open subsets will be useful in describing morphisms from P n (k). We describe the identification with affine space. Consider the map

Proof

: U L → An+1 (k) [a0 , . . . , an ] → (a0 /L(a0 , . . . , an ), . . . , an /L(a0 , . . . , an )) and write b0 , . . . , bn for the coordinates on affine space. The image of is the affine hyperplane V = {(b0 , . . . , bn ) : c0 b0 + · · · + cn bn = 1} ⊂ An+1 (k). Note that V An (k): If ci = 0 then the projection

: V → An (k) (b0 , . . . , bn ) → (b0 , . . . , bi−1 , bi+1 , . . . , bn ) has inverse

(b0 , . . . , bi−1 , bi+1 , . . . , bn ) → (b0 , . . . , bi−1 , 1 −

c j b j /ci , bi+1 , . . . , bn ).

j=i

Thus we obtain a bijective map ψ L = ( ◦ ) : U L → An (k) [a0 , . . . , an ] → (a0 /L , . . . , ai−1 /L , ai+1 /L , . . . , an /L).

148

PROJECTIVE GEOMETRY

This discussion might leave a nagging doubt in the reader’s mind: is ψ L compatible with the identifications ψi : Ui → An (k) used in the definition of projective space? Consider how these fit together over the intersections ψj

An (k)

UL ∩ U j ψL

An (k)

where we assume i < j for notational simplicity. The composition ϕ j (U L ∩ U j ) ||

ψ L ◦ψ −1 j

→ ρi j

ψ L (U L ∩ U j ) ||

An (k) \ {L = 0} → An (k) \ {b j = 0}, where L = L(b0 , . . . , bj−1 , 1, bj+1 , . . . , bn ), is given by (b0 /L , . . . , bi−1 /L , bi+1 /L , . . . , bj−1 /L , 1/L , bj+1 /L , . . . , bn /L).

These are birational, just as in the case of the distinguished affine open subsets.

A more formal approach, explicitly using the language of abstract varieties, is sketched in Exercise 9.22.

Remark 9.33

Let X ⊂ P n (k) be projective and L ∈ k[x0 , . . . , xn ] be a linear form such that L ∈ J (X ). Then

Corollary 9.34

U L ∩ X ⊂ U L An (k) is naturally an affine variety. A linear map φ : P n (k) P m (k) is a polynomial map induced by a linear transformation, i.e., there exists an (m + 1) × (n + 1) matrix A with entries in k such that

Definition 9.35

φ[x0 , . . . , xn ] = [a00 x0 + · · · + a0n xn , . . . , am0 x0 + · · · + amn xn ]. Fix a linear map τ : P n (k) P m (k) with matrix A. It defines a morphism P n (k) → P m (k) if A has trivial kernel. If X ⊂ P n (k) is projective then τ defines a morphism X → P m (k) if

Proposition 9.36 (Linear case)

J (X ) + L i := ai0 x0 + · · · + ain xn , i = 0, . . . , m is irrelevant.


149

Λ C

pΛC

P3

Figure 9.2

P2

Projection from a point in P3 .

A morphism τ : P n (k) → P n (k) induced by an invertible (n + 1) × (n + 1) matrix is called a projectivity. It suffices to prove the second assertion, which subsumes the first. Let V0 , . . . , Vm be the distinguished affine open subsets of P m (k) and U L i ⊂ P n (k) the corresponding affine open subsets. We tacitly ignore rows of A which are zero. Using the identifications of Proposition 9.32, τ induces a sequence of morphisms of affine varieties

Proof

τ

U L i → Vi , i = 0, . . . , m || || τi n A (k) → Am (k). Assuming that the entry ai, = 0, we can compute τi : ψ L−1,P n

(b0 , . . . , b−1 , b+1 , . . . , bn ) → [b0 , . . . , b−1 , (1− i

ψi,P m

τ

j=

ai j b j )/ai , b+1 , . . . , bn ]

→ [L 0 , . . . , L i−1 , 1, L i+1 , . . . , L m ] → (L 0 , . . . , L i−1 , L i+1 , . . . , L m)

where L r

= ar 0 b0 + · · · + ar −1 b−1 + ar 1 −

ai j b j /ai + ar +1 b+1 + · · · ar n bn

j=

for r = i. These are morphisms of affine spaces. These together define a morphism X → P m (k) provided the U L i , i = 0, . . . , m cover X . In the language of Section 9.8, {U L i ∩ X }i=0,...,m is an affine open covering of X . This is guaranteed by our irrelevance hypothesis: the linear equations L0 = L1 = . . . = Lm = 0 have no nontrivial solutions along X –even over extensions of k. Thus the U L i contain all of X .

150

PROJECTIVE GEOMETRY

Let ⊂ P n (k) be a linear subspace, i.e., the locus defined by the vanishing of linearly independent forms L 0 , . . . , L m for m < n. The polynomial map

Definition 9.37

π : P n (k) P m (k) [x0 , . . . , xn ] → [L 0 , . . . , L m ] is called projection from the subspace . 9.6.2 Veronese morphisms

We give another special instance of Proposition 9.29. Because of the importance of this example, we offer a precise description of its image: Proposition 9.38 (Veronese embedding)

The polynomial map

n+d ν(d) : P n (k) → P(k[x0 , x1 , . . . , xn ]d ) = P( d )−1 (k) α [x0 , . . . , xn ] → x = x0α0 . . . xnαn

all monomials with |α|=d

is a morphism. Its image is closed and defined by the equations zα zβ = zγ zδ

(9.2)

for all indices with α + β = γ + δ. Moreover, ν(d) is an isomorphism onto its image. This is called the Veronese embedding, after Giuseppe Veronese (1854–1917). Proof

The equations correspond to the identities xα xβ = xγ xδ,

α + β = γ + δ.

We show that ν(d) defines a morphism by analyzing it over each of the distinguished affine open subset Ui ⊂ P n (k). For notational simplicity we assume i = 0. Now U0 is mapped into the distinguished affine open subset n+d Ud0...0 = {z d0...0 = 0} ⊂ P( d )−1 (k),

so we must show that the induced map of affine spaces ν(d)0 : U0 → Ud0...0 is a morphism. Let y1 , . . . , yn and w α1 ...αn be coordinates on U0 and Ud0...0 respectively. The second set of variables is indexed by nonconstant monomials of degree ≤ d in n variables, i.e., the dehomogenizations of the x α other than x0d . The map ν(d)0 is given by (y1 , . . . , yd ) → (y α )0 0. These are instances of the equations (9.2). To complete the argument, we check that the projective variety defined by the equations (9.2) lies in the union of the n + 1 distinguished affine open subsets Ud0...0 ∪ U0d0...0 ∪ . . . U0...0d . (Then the local analysis above implies ν(d) is an isomorphism onto its image.) Suppose we have z d0...0 = z 0d0...0 = . . . = z 0...0d = 0. Iteratively applying the equations (9.2), we obtain αn α0 α1 z αd 0 ...αn = z d0...0 z 0d0...0 . . . z 0...0d

for each α. (This combinatorial deduction is left as an exercise.) It follows that z α0 ...αn = 0 for each α. We pause to flesh out some important special cases of Proposition 9.38:

152

PROJECTIVE GEOMETRY

The Veronese embeddings of P1 take the form

Example 9.39

φ : P1 (k) → P m (k) [x0 , x1 ] → x0m , x0m−1 x1 , . . . , x1m . The image is called the rational normal curve of degree m. It is defined by the 2 × 2 minors of the 2 × m matrix zm 0 z m−1 1 . . . z 1 m−1 . z m−1 1 z m−2 2 . . . z 0 m The degree-2 Veronese embedding takes the form

Example 9.40

n+2 φ : P n (k) → P( 2 )−1 (k) [x0 , x1 , . . . , xn ] → x02 , x0 x1 , x0 x2 . . . , xn−1 xn , xn2 .

all monomials of degree 2

The image is defined by the 2 × 2 minors of the symmetric (n + 1) × (n + 1) matrix M, where Mi j = M ji is the coordinate function mapped to xi x j . Proposition 9.38 gives a large number of affine open subsets in P n (k). Let F ∈ k[x0 , . . . , xn ] be homogeneous of degree d > 0. The

Proposition 9.41

open subset VF = {(x0 , . . . , xn ) ∈ P n (k) : F(x0 , . . . , xn ) = 0} ⊂ P n (k) naturally carries the structure of an affine variety. Proof

Express F=

cα x α

|α|=d

and write HF := {[z α ] : L F =

α

n+d cα z α = 0} ⊂ P( d )−1 (k)

for the corresponding hyperplane with complement

U F := [z α ] : cα z α = 0 . α

By Proposition 9.32, U F is naturally affine space with coordinate ring

k[bα ]/ cα bα − 1. α


153

The Veronese map ν(d) identifies VF with the intersection U F ∩ ν(d)(P n ). Proposition 9.38 implies ν(d)(P n ) is cut out by homogeneous equations, so n+d U F ∩ ν(d)(P n ) ⊂ U F A( d )−1

is defined by polynomial equations and thus is affine. Remark 9.42

The coordinate ring of U F has an alternate description k[U F ] k[z α ][L −1 F ]degree 0 ,

the quotients G/L NF with G homogeneous of degree N in the z α . With a bit more work, we could compute analogously k[VF ] = (k[x0 , . . . , xn ][1/F])degree 0 , the homogeneous fractions in k[x0 , . . . , xn ][1/F] of degree zero. Example 9.43

Let F = x12 + 4x1 x2 − 3x22 . Then we can realize VF ⊂ P2 (k) as

the affine variety 2 (y02 , y11 , y20 ) : y11 = y02 y20 , y20 + 4y11 − 3y02 = 1 . Let X = {(x0 , x1 , x2 ) : x03 + x13 + x23 = 0} ⊂ P2 (C) be a plane curve. The rule from Example 9.28 does define a morphism on X

Example 9.44

φ : X → P1 (C) [x0 , x1 , x2 ] → x02 + x12 , x12 + x22 . We have φ −1 [0, 0] = {[1, i, 1], [1, −i, 1], [−1, i, 1], [−1, −i, 1]} ⊂ P2 (k), which is disjoint from X . 9.6.3 Proof of Propositions 9.29 and 9.30

Every polynomial rational map φ : P n (k) P m (k) can be factored n+d τ ν(d) P n (k) → P( d )−1 (k) P m ,

where ν(d) is the Veronese embedding and τ is linear. Indeed, writing each coordinate function

φi = ci;α x α , deg(α) = d, α

154

PROJECTIVE GEOMETRY

τ arises from the (m + 1) × tions 9.38 and 9.36.

9.7

n+d d

matrix (ci;α ). Our results follow from Proposi-

Products The product P m (k) × P n (k) is a projective variety. It can be realized as the locus in Pmn+m+n (k) where the 2 × 2 minors of

Proposition 9.45 (Segre embedding)

⎛

z 00 ⎜ z 10 ⎜ ⎝ ... z m0

z 01 z 11 ... z m1

... ... ... ...

⎞ z 0n z 1n ⎟ ⎟ ... ⎠ z mn

(9.3)

vanish. Here we regard Pmn+m+n (k) as the projective space of all (m + 1) × (n + 1) matrices (z i j ). This is called the Segre embedding of P m (k) × P n (k) into Pmn+m+n (k), after Corrado Segre (1863–1924). Our analysis here presumes the abstract variety structure on the product P (k) × P (k) (cf. Proposition 9.52). Our first task is to construct a morphism

Proof m

n

φ : P m (k) × P n (k) → P(m+1)(n+1)−1 (k). This is given by the rule [x0 , . . . , xm ] × [y0 , . . . , yn ] → [x0 y0 , x0 y1 , . . . , x0 yn , x1 y0 , . . . , x1 yn , x2 y0 , . . .], which is well-defined on projective equivalence classes. Consider the distinguished affine open subsets Vi = {[x0 , . . . , xm ] : xi = 0} ⊂ P m (k), W j = {[y0 , . . . , yn ] : y j = 0} ⊂ P n (k), and Ui j = {[z 00 , . . . , z mn ] : z i j = 0} ⊂ Pmn+m+n (k). Note that φ(Vi × W j ) ⊂ Ui j for each i and j. The restriction of φ to V0 × W0 takes the form V0 × W0 → U00 [1, x1 /x0 , . . . , xm /x0 ] , [1, y1 /y0 , . . . , yn /y0 ] → [xi y j /x0 y0 ]. If a1 , . . . , am and b1 , . . . , bn are coordinates on V0 Am (k) and W0 An (k), φ can be expressed φ00 : Am (k) × An (k) → Amn+m+n (k) (a1 , . . . , am ,b1 , . . . , bn ) → (b1 , b2 , . . . , bn , a1 , a1 b1 , . . . , a1 bn , a2 , a2 b1 ,. . . , am bn ). This is a morphism of affine varieties.

9.7 PRODUCTS

155

We claim that image(φ) is closed and

P m (k) × P n (k) → image(φ). Fix coordinates on U00 : c01 , . . . , c0n , c11 , . . . , c1n , . . . , cmn such that ∗ ∗ ∗ φ00 c0 j = b j , φ00 ci0 = ai , φ00 ci j = ai b j ,

i = 1, . . . , m, j = 1, . . . , n.

The image of φ00 is the closed set defined by the relations ci j = ci0 c0 j , which is isomorphic to affine space Am+n (k) with coordinates c10 , . . . , cm0 , c01 , . . . , c0n . (Indeed, φ00 has a left inverse: project onto the variables corresponding to the a j and bi .) Repeating this over each of the distinguished affine opens, we conclude that each local realization φi j : Vi × W j → Ui j has closed image and is an isomorphism onto its image. It remains to extract homogeneous equations for the image. Homogenizing ci j = ci0 c j0 yields z i j z 00 = z i0 z 0 j . Working over all the distinguished affine open sets, we obtain the relations z i j z h = z i z h j ,

i, h = 0, . . . , m, j, = 0, . . . , n,

which are the 2 × 2 minors of our matrix.

Identify the (m + 1) × (n + 1) matrices (z i j ), up to scalar multiplication, with Pmn+m+n (k). The matrices of rank 1 form a closed subset R1 ⊂ Pmn+m+n (k), defined by the 2 × 2 minors of (z i j ). The image of a rank-1 matrix (z i j ) is a one-dimensional subspace of k m+1 , thus a point in P m (k). The kernel is a codimension-1 linear subspace of k n+1 , defined by a linear form

Remark 9.46

ker(z i j ) = {(w 0 , . . . , w n ) ∈ k n+1 : a0 w 0 + · · · + an w n = 0}, where (a0 , . . . , an ) is proportional to any of the nonzero rows of (z i j ). We regard [a0 , . . . , an ] ∈ P n (k), the projective space associated to the vector space dual to k n+1 . This gives us a function R1 → P m (k) × P n (k) (z i j ) → (image(z i j ), linear form vanishing on ker(z i j )). Since there is a unique rank-1 matrix with prescribed kernel and image, this is a bijection. We can now address one complication that arises in the study of morphisms of projective varieties. Consider projection onto the first factor π1 : P1 (k) × P1 (k) → P1 (k) ([x0 , x1 ], [y0 , y1 ]) → [x0 , x1 ],

156

PROJECTIVE GEOMETRY

which is definitely a morphism. And we have just seen how P1 (k) × P1 (k) can be realized as the surface {[z 00 , z 01 , z 10 , z 11 ] : z 00 z 11 = z 01 z 10 } ⊂ P3 (k). Does π1 come from a polynomial map P3 (k) → P1 (k)? Definitely not! Any degree-d polynomial in the z i j corresponds to a polynomial homogeneous of degree d in both {x0 , x1 } and {y0 , y1 }; the degree in the xi equals the degree in the yi . However, π1 cannot be represented with such polynomials. The polynomial rational maps P3 (k) P1 (k) best approximating π1 are φ : P3 (k) P1 (k) [z 00 , z 01 , z 10 , z 11 ] → [z 00 , z 10 ], φ : P3 (k) P1 (k) [z 00 , z 01 , z 10 , z 11 ] → [z 01 , z 11 ]. Note that φ agrees with π1 away from the line {z 00 = z 10 = 0} = P1 (k) × [0, 1] ⊂ P1 (k) × P1 (k) ⊂ P3 (k); φ agrees with π1 away from the line {z 01 = z 11 = 0} = P1 (k) × [1, 0] ⊂ P1 (k) × P1 (k) ⊂ P3 (k).

9.8

Abstract varieties We would like to endow P n (k) with the structure of a variety, compatible with our decomposition (9.1) as a union of affine spaces. To achieve this we will need a more flexible definition of algebraic varieties, going beyond the affine examples we have studied up to this point. This section can be omitted on first reading.

9.8.1 Graphs of birational maps

Let ρ : V W be a birational map and ρ ⊂ V × W the closure of its graph. Let pV : ρ → V and pW : ρ → W be the morphisms induced by the projections. Let Iρ ⊂ k[V ] denote the indeterminacy ideal of ρ and Iρ −1 ⊂ k[W ] the indeterminacy ideal of its inverse. Consider the intersection of the graphs of ρ and ρ −1 Uρ = {(v, w) : ρ defined at v, ρ −1 defined at w}. We can regard this as an open subset of both V and W . By Proposition 8.37, we have −1 Uρ = ρ \ pV−1 V (Iρ ) ∪ pW V (Iρ −1 ) . We are interested in whether Uρ = ρ , i.e., whether the indeterminacy pV−1 V (Iρ ) −1 or pW V (Iρ −1 ) actually meets ρ . This might be sensitive to the field – indeterminacy may be apparent only after base extension (see Example 3.49.) However, if k is

9.8 ABSTRACT VARIETIES

157

algebraically closed then Proposition 8.40 (or a direct argument using the Nullstellensatz) implies that Uρ = ρ if and only if ∗ I (ρ ) + ( pV∗ Iρ ∩ pW Iρ −1 ) = k[V × W ].

We say then that ρ : V W satisfies the closed graph condition. In this case, Uρ is naturally an affine variety. Example 9.47

The birational map ρ : A1 (k) A1 (k) x → 1/x

satisfies the closed graph condition. We have I (ρ ) = x z − 1, Iρ = x, and Iρ −1 = z, so that I (ρ ) + (k[x, z]Iρ ∩ k[x, z]Iρ −1 ) = 1 − x z, x z = k[x, z]. The birational morphism ρ : A2 (k) A2 (k) (x1 , x2 ) → (x1 , x1 x2 ) does not satisfy the closed graph condition. Here I (ρ ) = y1 − x1 , y2 − x1 x2 , Iρ = k[x1 , x2 ], and Iρ −1 = y1 , so we have y1 − x1 , y2 − x1 x2 , y1 = k[x1 , x2 , y1 , y2 ]. 9.8.2 Coverings of affine varieties

For pedagogical reasons, we introduce the ideas behind abstract varieties in the familiar context of affine varieties, where they are more transparent. Let V be an affine variety. Choose g0 , . . . , gn ∈ k[V ] and write Ui = {v ∈ V : gi (v) = 0} for each i. Suppose that g0 , . . . , gn = k[V ], so there exist h 0 , . . . , h n ∈ k[V ] with h 0 g0 + · · · + h n gn = 1. In particular, for each v ∈ V we have gi (v) = 0 for some i and V = U0 ∪ U1 . . . ∪ Un . If k is algebraically closed then the fact that V can be expressed as a union of the Ui s implies that g0 , . . . , gn = k[V ]; this follows from the Nullstellensatz (Theorem 7.5). As in the proof of Proposition 3.47, we endow each open subset Ui with the structure of an affine variety. Let Vgi := {(v, z) : gi (v)z = 1} ⊂ V × A1 (k)

158

PROJECTIVE GEOMETRY

so that projection onto V maps Vgi bijectively onto Ui . The collection {Vgi }i=0,...,n is called an affine open covering of V . The intersections Ui1 ∩ . . . ∩ Uir also are naturally affine varieties; they correspond to the affine open subsets Vgi1 −gir . Here is the crucial property of affine open coverings: Let V and W be affine varieties. A morphism φ : V → W is equivalent to the data of an affine open covering {Vgi }i=0,...,n and a collection of morphisms {φi : Vgi → W, i = 0, . . . , n} satisfying the compatibility φi |Vgi g j = φ j |Vgi g j .

Theorem 9.48

Proof

Given φ : V → W , we obtain φi by composing Vgi → V → W

where the first arrow is the standard projection morphism. These are compatible on the intersections Vgi g j by construction. Conversely, suppose we are given a collection {φi } as above. By Proposition 3.31, we get a collection of k-algebra homomorphisms φi∗ : k[W ] → k[V ][gi−1 ]. The morphism Vgi g j → Vgi corresponds to the localization homomorphism k[V ][gi−1 ] → k[V ][(gi g j )−1 ]. By assumption the compositions φi∗ k[W ] → k[V ] gi−1 → k[V ][(gi g j )−1 ]

φ ∗j → k[V ][(gi g j )−1 ] k[W ] → k[V ] g −1 j

are equal. We construct a unique k-algebra homomorphism ψ : k[W ] → k[V ] such that for each f ∈ k[W ] and i = 0, . . . , n we have ψ( f ) = φi∗ f in k[V ][gi−1 ]. Our desired φ : V → W is the unique morphism with φ ∗ = ψ. Lemma 9.49

Let R be a ring and g0 , . . . , gn ∈ R such that h 0 g0 + h 1 g1 + · · · + h n gn = 1

for some h 0 , . . . , h n ∈ R. Then for each N > 0 there exist h 0 , . . . , h n such that h 0 g0N + h 1 g1N + · · · + h n gnN = 1.

Proof of lemma:

Taking N (n + 1) powers and expanding gives

Pe0 ...en (h 0 , . . . , h n )g0e0 . . . gnen = 1

e0 +···+en =N (n+1)


159

where Pe0 ...en (h 0 , . . . , h n ) is a suitable polynomial in the h i . For each term, one of the ei ≥ N , i.e., each summand is divisible by giN for some i. Regrouping terms in the summation gives the result. ˆ f ) ∈ k[V ] with First, if ψ( f ) exists it must be unique: Suppose we have another ψ( −1 ˆ ˆ f)− ψ( f ) = ψ( f ) ∈ k[V ][gi ] for each i. Then there exists an N such that giN (ψ( ψ( f )) = 0 for each i. It follows that ˆ f ) − ψ( f ) = ψ(

n

ˆ f ) − ψ( f )) = 0. h i giN ((ψ(

i=0

Second, observe that this uniqueness forces ψ to be a homomorphism: Given f 1 , f 2 ∈ k[W ], ψ( f 1 ) = φi∗ ( f 1 ) and ψ( f 2 ) = φi∗ ( f 2 ) in k[V ][gi−1 ] imply that ψ( f 1 ) + ψ( f 2 ) = φi∗ ( f 1 + f 2 ). Thus ψ( f 1 + f 2 ) = ψ( f 1 ) + ψ( f 2 ). A similar argument proves ψ( f 1 f 2 ) = ψ( f 1 )ψ( f 2 ). It suffices then to write a formula for ψ and verify the required properties. Given f ∈ k[W ], choose N such that giN φi∗ ( f ) ∈ k[V ] for each i and (gi g j ) N φi∗ ( f ) = (gi g j ) N φ ∗j ( f ) for each i, j. Write ψ( f ) =

n

h j g Nj φ ∗j ( f )

j=0

so that n

N ∗ j=0 h j (g j gi ) φ j ( f ) n = j=0 h j (g j gi ) N φi∗ ( f ) n N giN φi∗ ( f ) = j=0 h j g j

giN ψ( f ) =

= 1 · giN φi∗ ( f ) which means that ψ( f ) = φi∗ ( f ) in k[V ][gi−1 ].

How are the various affine open subsets Vgi related? Suppose that gi , g j ∈ k[V ] are not zero divisors so that k Vgi = k(V ) = k Vg j . In particular, Vgi and Vg j are birational, and Corollary 3.46 yields inverse rational maps ρi j : Vg j Vgi ,

ρ ji : Vgi Vg j .

The birational maps ρi j are called the gluing maps.

160

PROJECTIVE GEOMETRY

The birational maps ρi j satisfy the closed-graph property. Since Vgi g j ⊂ ρi j ⊂ Vgi × Vg j it suffices to show that Vgi g j is closed in the product. If πi : Vgi → V and π j : Vg j → V are the standard projections then (πi , π j )−1 V = {(v 1 , v 2 ) : πi (v 1 ) = π j (v 2 )} ⊂ Vgi × Vg j ⊂ (V × A1 (k))2 is closed by Exercise 3.10. This is equal to {(v, z i , z j ) ∈ V × A1 (k) × A1 (k) : z i gi (v) = z j g j (v) = 1}, i.e., the affine variety with coordinate ring k[V ][gi−1 , g −1 j ]. The isomorphism of kalgebras ∼

−1 k[V ][gi−1 , g −1 j ] → k[V ][(gi g j ) ]

induces an isomorphism of affine varieties ∼

Vgi g j → (πi , π j )−1 V . 9.8.3 Definition of abstract varieties

An abstract variety X consists of a collection of affine varieties U0 , . . . , Un and birational maps ρi j : U j Ui satisfying the closed-graph property and the compatibility conditions ρi j ◦ ρ j = ρi for all i, j, ∈ {0, . . . , n}. We regard X as the quotient of the disjoint union U0 . . . Un under the equivalence relation ≈ generated by the following ‘gluing data’: Given u i ∈ Ui and u j ∈ U j , u i ≈ u j if ρi j is defined at u j and u i = ρi j (u j ). The ρi j are called the gluing maps. The intersection Ui ∩ U j ⊂ X therefore has the structure of the affine variety Ui j := ρi j . The collection {Ui }i=0,...,n is called an affine open covering of X . We say that Z ⊂ X is closed if each intersection Z ∩ Ui is closed in Ui . Let V be affine and g0 , . . . , gn ∈ k[V ] as in § 9.8.2. Then the collection of open affines

Example 9.50

Ui = {v ∈ V : gi (v) = 0} makes V into an abstract variety.


161

A refinement {Wi; } of a covering {Ui } is a union of open coverings {Wi; }=1,...,m i of each Ui , as described in §9.8.2. There are induced birational gluing maps among the Wi; . Let X be an abstract variety and W an affine variety. A morphism φ : X → W is specified over an affine open covering {Ui } of X by a collection of morphisms φi : Ui → W compatible with the gluing maps ρi j : U j → Ui , i.e., φ j = φi ◦ ρi j , for each i, j. The compatibility condition is equivalent to stipulating that φi |Ui j = φ j |Ui j , for each i, j. Theorem 9.48 shows that our new definition is consistent with our orginial definition when X = V is affine and {Ui = Vgi } arises as in § 9.8.2. It also guarantees our definition respects refinements of affine open coverings. We do not distinguish between a morphism specified over a covering and the morphisms arising from refinements of that covering. A morphism of abstract varieties φ : X → Y is specified by the following data: affine open coverings {Ui }i=0,...,n and {V j } j=0,...,m with gluing maps ρia : Ua Ui and ξ jb : Vb V j for X and Y respectively, and a collection of morphisms

φ j(i),i : Ui → V j(i)

i=0,...,n

,

with j(i) ∈ {0, . . . , m}, such that the compatibility condition φ j(i),i ◦ ρia = ξ j(i) j(a) ◦ φ j(a),a holds for each i, a ∈ {0, . . . , n}. Equivalently φ j(i),i Uia = ξ j(i), j(a) ◦ φ j(a),a Uia holds on the interesections Uia = Ui ∩ Ua . If we fix coverings {Ui } and {V j } of X and Y then most φ : X → Y cannot be realized by compatible collections of morphisms {φ j(i),i : Ui → V j(i) }. As we saw in our analysis of morphisms of projective spaces, the choice of covering must take into account the geometry of the morphism.

Remark 9.51

We define rational maps of abstract varieties analogously. The formalism of abstract varieties is quite useful for constructing products: Let X and Y be abstract varieties. Then the product X × Y has a natural structure as an abstract variety and admits natural projection morphisms π1 : X × Y → X and π2 : X × Y → Y .

Proposition 9.52

Let {Ui }i=0,...,m and {V j } j=0,...,n be coverings of X and Y respectively by affine open subsets, and ρia : Ua Ui and ξ jb : Vb V j the birational

Sketch proof

162

PROJECTIVE GEOMETRY

gluing maps. The product X × Y is covered by the affine varieties {Ui × V j } with gluing maps (ρia , ξ jb ) : Ua × Vb Ui × V j . The projections π1 : Ui × V j → Ui ,

π2 : Ui × V j → V j

are clearly compatible with the gluing maps, and thus define the desired morphisms.

9.9

Exercises

9.1

Show that an ideal J ⊂ k[x0 , . . . , xn ] is homogeneous if, and only if, for each f ∈ J the homogeneous pieces of f are all in J . For a = [a0 , . . . , an ] ∈ P n (k), show that

9.2

J (a) = xi a j − x j ai i, j=0,...,n . 9.3

Let Z = P1 (k) P1 (k) P1 (k) be the disjoint union of three copies of P1 (k). Show that Z X × P1 (k) where X = { p0 , p1 , p2 } ⊂ P1 (k). Realize Z explicitly as a projective variety and find its homogeneous equations. Consider the disjoint union W = P1 (k) P1 (k) . . . P1 (k) . n times

9.4 9.5

Explain how W can be realized as a closed subset of P3 (k). Let I, J ⊂ k[x0 , . . . , xn ] be irrelevant ideals. Show that I ∩ J and I + J are also irrelevant. Consider the polynomial map [x0 , x1 , x2 , x3 ] → [x0 x1 , x0 x2 , x0 x3 , x1 x2 , x1 x3 , x2 x3 ]. (a) Does this induce a well-defined morphism φ : P3 (Q) → P5 (Q)? (b) Consider the hypersurface X = [x0 , x1 , x2 , x3 ] : x02 + x12 + x22 + x32 = 0 ⊂ P3 (C).

9.6

Does our polynomial map induce a well-defined morphism X → P5 (C)? Consider the homogeneous ideal

J = x02 x1 , x13 , x1 x22 .

9.9 EXERCISES

9.7

163

(a) Let Ii , i = 0, 1, 2 denote the dehomogenization of J with respect to xi . Compute each Ii . (b) Let Ji be the homogenization of Ii with respect to xi . Compute each Ji . (c) Compute the intersection J := J0 ∩ J1 ∩ J2 . Show that J J . (d) Show that (J : x0 , x1 , x2 ) = J . Describe the irreducible components of the projective variety X = {[x0 , x1 , x2 , x3 ] : x0 x1 − x2 x3 = x0 x2 − x1 x3 = 0} ⊂ P3 (k).

9.8

Hint: You may find it easier to work first in the distinguished affine neighborhoods. Consider the ideal I = y2 − y12 , y3 − y1 y2 , . . . , yn − y1 yn−1 = yi+1 − y1 yi , i = 1, . . . , n − 1. Show that the homogenization of I is generated by the 2 × 2 minors of the matrix

9.9

9.10

x0 x1

x1 x2

... ...

xn−1 xn

.

Compare this with Examples 1.5 and 9.39 and Exercise 1.3. Consider the rule φ : [s, t] → [s 4 , s 2 t 2 , t 4 ]. (a) Show that φ defines a morphism P1 (k) → P2 (k). (b) Compute the image φ(P1 (k)) ⊂ P2 (k) and its closure φ(P1 (k)). (c) Show that φ is not dominant. Consider the polynomial rational map φ : P1 (k) P2 (k) [x0 , x1 ] → x03 , x02 x1 , x0 x12 . Show this is well-defined on the distinguished open set U0 ⊂ P1 (k) and there is a morphism φˆ : P1 (k) → P2 (k)

9.11

ˆ 0 = φ|U0 . such that φ|U Let f = ad x d + · · · + a0 and g = be x e + · · · + b0 be nonzero polynomials in x. Consider the corresponding points [ f ] = [a0 , a1 , . . . , ad ] ∈ Pd (k) and [g] = [b0 , b1 , . . . , be ] ∈ Pe (k). Show that {([ f ], [g]) : Res( f, g) = 0} ⊂ Pd (k) × Pe (k)

9.12

is a well-defined closed subset. Let X ⊂ P n (k) be closed and ⊂ P n (k) a finite set disjoint from X . Show there exists an affine open subset V with ⊂ V ⊂ (P n (k) \ X ).

164

PROJECTIVE GEOMETRY

9.13

Consider the complex plane curves C1 = [x0 , x1 , x2 ] : x02 x2 = x13 and C2 = [x0 , x1 , x2 ] : x0 x22 = x0 x12 + x13 . Show that each of these is the image of the rational normal cubic curve X = ν(3)(P1 ) ⊂ P3 (C)

9.14

under projection from a point ∈ P3 (C) not on that curve. Our discussion of projective geometry ignores its historical origins in the study of parallelism. Here we partially fill this gap. A line L ⊂ P2 (k) is defined L = {[x0 , x1 , x2 ] : p0 x0 + p1 x1 + p2 x2 = 0} where ( p0 , p1 , p2 ) = 0. Note that ( p0 , p1 , p2 ) and λ( p0 , p1 , p2 ), λ ∈ k ∗ , define the same line. Thus the lines in P2 (k) are also parametrized by a projective plane, the dual projective plane Pˇ 2 (k). (a) Show that the incidence correspondence W = {[x0 , x1 , x2 ], [ p0 , p1 , p2 ] : p0 x0 + p1 x1 + p2 x2 = 0} ⊂ P2 (k) × Pˇ 2 (k) is closed. Check that the projections p : W → P2 (k),

pˇ : W → Pˇ 2 (k)

have projective lines as fibers, i.e., p −1 ([a0 , a1 , a2 ]) P1 (k). (Translation: the lines through a point in the plane form a projective line.) (b) Show that the open subset U = Pˇ 2 \ {[1, 0, 0]} corresponds to lines in the affine plane A2 (k) = U0 = {x0 = 0} ⊂ P2 (k). (c) Show that Y = {[x0 , x1 , x2 ], [ p0 , p1 , p2 ], [q0 , q1 , q2 ] : p0 x0 + p1 x1 + p2 x2 = 0, q0 x0 + q1 x1 + q2 x2 = 0} ⊂ P2 (k) × Pˇ 2 (k) × Pˇ 2 (k) is closed. (d) Show that the open subset V = {[ p0 , p1 , p2 ], [q0 , q1 , q2 ] : p1 q2 − p2 q1 = 0} ⊂ U × U parametrizes pairs of non-parallel lines in the affine plane. (e) Let q : Y → P2 (k) and qˇ : Y → Pˇ 2 (k) × Pˇ 2 (k) be the projection morphisms. ˇ −1 (U0 )) contains V . (Translation: any two non-parallel lines in Show that q(q the affine plane intersect somewhere in the affine plane.)

9.9 EXERCISES

9.15

9.16

165

ˇ ) = Pˇ 2 (k) × Pˇ 2 (k). (Translation: any two lines in the projective (f) Show that q(Y plane intersect.) (g) Let 1 , . . . , r ⊂ A2 (k) be a collection of parallel lines and L 1 , . . . , L r their closures in P2 (k). Show that L 1 , . . . , L r share a common point s ∈ H0 = {x0 = 0}, the line at infinity. Moreover, each point s ∈ H0 arises in this way. (Translation: the points on the line at infinity correspond to equivalence classes of parallel lines in the affine plane.) Let X be a projective variety. Show that X is isomorphic to a variety Y ⊂ P N (k) defined by quadratic equations. Hint: Given X ⊂ P n defined by homogeneous equations of n+d degree ≤ d, analyze the equations satisfied by Y = ν(d)(X ) in P( d )−1 (k). It is quite fruitful to consider real morphisms φ : P n (R) → P m (R) whose image happens to lie in a distinguished open subset. Consider the following version of the 2-Veronese morphism: ϕ : P1 (R) → P2 (R) [x0 , x1 ] → x02 + x12 , x02 − x12 , 2x0 x1 . Show that the image satisfies the equation z 12 + z 22 = z 02 and verify that ϕ(P1 (R)) ⊂ U0 . Conclude that there is a bijection between the real projective line and the unit circle C = (y1 , y2 ) : y12 + y22 = 1 .

9.17

Challenge: Show that P1 (R) and C are not isomorphic as varieties, by proving there exists no non-constant morphism P1 (R) → A1 (R). (Steiner Roman surface) This is a continuation of Exercise 9.16. Consider the polynomial map φ : P2 (R) P3 (R) [x0 , x1 , x2 ] → x02 + x12 + x22 , x1 x2 , x0 x2 , x0 x1 . (a) Show that φ is a morphism and φ(P2 (R)) ⊂ U0 . (b) Verify that image(φ) satisfies the quartic z 12 z 22 + z 12 z 32 + z 22 z 32 = z 0 z 1 z 2 z 3 . (c) Writing V = (y1 , y2 , y3 ) : y12 y22 + y12 y32 + y22 y32 = y1 y2 y3 ⊂ A3 (R)

166

PROJECTIVE GEOMETRY

Figure 9.3

Steiner Roman surface.

we have a well-defined function ϕ : P2 (R) → V.

9.18

For each point v ∈ V , determine #{ p ∈ P2 (R) : ϕ( p) = v}. Hint: Analyze carefully what happens along the coordinate axes. The surface V ⊂ R3 is known as the Steiner Roman surface, after Jakob Steiner (1796–1863). It is a useful tool for visualizing the real projective plane. (a) Working from the definitions, show that P1 (k) × P1 (k) × P1 (k) is an abstract variety. (b) Verify that the rule φ : P1 (k) × P1 (k) × P1 (k) → P7 (k) taking ([s0 , s1 ], [t0 , t1 ], [u 0 , u 1 ]) to [s0 t0 u 0 , s0 t0 u 1 , s0 t1 u 0 , s0 t1 u 1 , s1 t0 u 0 , s1 t0 u 1 , s1 t1 u 0 , s1 t1 u 1 ] gives a well-defined morphism of varieties. (c) Show that the image of φ is closed in P7 (k) and write down equations for the image.

9.9 EXERCISES

9.19

167

(d) Check that φ is an isomorphism onto its image and conclude that P1 (k) × P1 (k) × P1 (k) is projective. Let X be a projective variety. (a) Show that the diagonal map :X → X×X x → (x, x) is a morphism. Its image is denoted X . (b) Let X = P n (k) and fix coordinates x0 , x1 , . . . , xn and y0 , y1 , . . . , yn on P n (k) × P n (k). Show that I (P n (k) ) = xi y j − x j yi i, j=0,...,n .

9.20

(c) For general X , show that X is closed in X × X and isomorphic to X . (d) Repeat parts (a) and (c) for an arbitrary abstract variety. Hint: Let {Ui } be a covering for X with gluing maps ρi j : U j Ui . Verify that X ∩ (Ui × U j ) corresponds to the graph of ρi j and apply the closed graph property. For each m, n > 0, show that there exists no constant morphism φ : P n (C) → Am (C). Hints: (a) Reduce to the m = 1 case: Suppose that for each projection πi : Am (C) → A1 (C) the composition πi ◦ φ is constant. Then φ is constant. (b) Reduce to the n = 1 case: Suppose that φ : P n (C) → A1 (C) is constant along each line ⊂ P n (C). Then φ is constant. (c) Express C[A1 ] = C[t]. Show there exists a collection of homogeneous forms {F0 , . . . , Fm } ⊂ C[x0 , x1 ]degree d such that F0 , . . . , Fm is irrelevant and φ ∗ t|VFi ∈ C[x0 , x1 ] Fi−1 degree 0 for each i (cf. Proposition 9.41). (d) Suppose F1 , F2 ∈ C[x0 , x1 ] are homogeneous of degree d with greatest common divisor G. Suppose that h ∈ C(x0 , x1 ) is a rational function expressible in two different forms h = H1 F1N , h = H2 F2N ,

9.21

where N > 0 and H1 and H2 are homogeneous of degree N d in C[x0 , x1 ]. Show that h = H/G for some homogeneous H with deg(H ) = deg(G). Let V be an affine variety over an algebraically closed field, and {gi }i∈I ∈ k[V ] a possibly infinite collection of elements such that V = ∪i∈I Ui ,

Ui = {v ∈ V : gi (v) = 0}.

168

PROJECTIVE GEOMETRY

9.22

Show that there exists a finite collection of indices i(0), . . . , i(n) such that {Vgi( j) } j=0,...,n is an affine open covering of V . In other words, every affine open covering admits a finite subcovering; this property is known as quasi-compactness. Recall the notation of Proposition 9.32. Consider the map φ L−1 : An (k) → U L ⊂ P n (k).

9.23

Show this defines a morphism An (k) → P n (k), using Theorem 9.48 and the definition of a morphism to an abstract variety. Recall that the birational morphism ρ

U0 := A2 (k) → U1 := A2 (k) (x1 , x2 ) → (x1 , x1 x2 ) with inverse ρ −1 (y1 , y2 ) = (y1 , y2 /y1 ) does not satisfy the closed graph condition. Show that the intersection U01 of the graphs of ρ and ρ −1 is nonetheless an affine variety with coordinate ring k[U01 ] = k[y1 , y2 ] y1−1 k[x1 , x2 ] x1−1 . Observe that U01 is naturally an affine open subset of both U0 and U1 . Describe the quotient of U0 U1 obtained by gluing along U01 : given u 0 ∈ U0 and u 1 ∈ U1 , we have u 0 ≈ u 1 if u 0 , u 1 ∈ U01 and ρ(u 0 ) = u 1 . Is this an affine variety? A projective variety?

10 Projective elimination theory

What is projective geometry good for? We focus on a key property of projective varieties that differentiates them from affine varieties: the image of a projective variety under a morphism is always closed. Recall that the image of a morphism of affine varieties is not closed. For example, consider the variety V = {(x, y) : x y = 1} ⊂ A2 (k) and the projection map π2 : A2 (k) → A1 (k) (x, y) → y. The image π2 (V ) = {y ∈ A1 (k) : y = 0} is not closed. This phenomenon complicates description of the image of a morphism. Of course, we have developed algorithms for computing the closure of the image of a map of affine varieties in Chapter 4. Projective geometry allows us to change the problem slightly, so that the image of the morphism becomes closed. We introduce the method for our archetypal example. Regard V ⊂ A1 (k) × A1 (k) ⊂ P1 (k) × A1 (k), where the affine factor corresponding to the variable x is completed to the projective line. The projective closure V ⊂ P1 (k) × A1 (k) is obtained by homogenization V = {([x0 , x1 ], y) : x1 y = x0 }. 169

170

PROJECTIVE ELIMINATION THEORY

We still have a projection map π2 : P1 (k) × A1 (k) → A1 (k) ([x0 , x1 ], y) → y, but now π2 (V ) = A1 (k). Indeed, the point added at infinity is mapped to the origin, π2 (([0, 1], 0) = 0.

10.1

Homogeneous equations revisited A polynomial F ∈ k[x0 , . . . , xn , y1 , . . . , ym ] is homogeneous of degree d in x0 , . . . , xn if F= x α h α (y1 , . . . , ym ), |α|=d

where the h α ∈ k[y1 , . . . , ym ]. Regard f ∈ k[w 1 , . . . , w n , y1 , . . . , ym ] as a polynomial in w 1 , . . . , w n with coefficients in k[y1 , . . . , ym ] of degree d. The homogenization of f relative to w 1 , . . . , w n is defined as F(x0 , . . . , xn , y1 , . . . , ym ) = x0d f (x1 /x0 , . . . , xn /x0 , y1 , . . . , ym ). An ideal J ⊂ k[x0 , . . . , xn , y1 , . . . , ym ] is homogeneous in x0 , . . . , xn if it can be generated by polynomials that are homogeneous in x0 , . . . , xn . Given I ⊂ k[w 1 , . . . , w n , y1 , . . . , ym ], the homogenization of I relative to w 1 , . . . , w n is the ideal J ⊂ k[x0 , . . . , xn , y1 , . . . , ym ] generated by homogenizations of elements in I . A monomial order > on k[w 1 , . . . , w n , y1 , . . . , ym ] is graded relative to w 1 , . . . , w n if it is compatible with the partial order induced by degree in the wvariables, i.e., w α y γ > w β y δ whenever |α| > |β|. We have the following straightforward generalization of Theorem 9.6: Let I ⊂ k[w 1 , . . . , w n , y1 , . . . , ym ] be an ideal and J ⊂ k[x0 , . . . , xn , y1 , . . . , ym ] its homogenization relative to w 1 , . . . , w n . Suppose that f 1 , . . . , fr is a Gröbner basis for I with respect to some order > graded relative to w 1 , . . . , w n . Then the homogenizations of f 1 , . . . , fr relative to w 1 , . . . , w n generate J . Proposition 10.1

For each F ∈ k[x0 , . . . , xn , y1 , . . . , ym ] homogeneous in x0 , . . . , xn of degree d, we have F(λx0 , . . . , λxn , y1 , . . . , ym ) = λd F(x0 , . . . , ym ) so the locus where F vanishes is well-defined in Pn (k) × Am (k). Let X (F) := {([x0 , . . . , xn ], y1 , . . . , ym ) ∈ Pn (k) × Am (k) : F(x0 , . . . , ym ) = 0}

10.2 PROJECTIVE ELIMINATION IDEALS

171

denote the corresponding hypersurface. More generally, for any ideal J ⊂ k[x0 , . . . , xn , y1 , . . . , ym ] homogeneous in x0 , . . . , xn , we have the closed subset X (J ) ⊂ Pn (k) × Am (k). For each subset S ⊂ Pn (k) × Am (k), write J (S) = {F ∈ k[x0 , . . . , xn , y1 , . . . , ym ] : F(s) = 0 for each s ∈ S}. We would like to find equations for closed subsets of Pn (k) × Am (k). When m = 0 we have already addressed this problem: Theorem 9.14 implies each closed X ⊂ Pn (k) is given by a homogeneous ideal. The general picture is very similar: Proposition 10.2

If S ⊂ Pn (k) × Am (k) then J (S) ⊂ k[x0 , . . . , xn , y1 , . . . , ym ]

is homogeneous and X (J (S)) = S. Moreover, J (S) can be computed effectively from the local affine equations vanishing along S. Consider the distinguished subsets of Pn (k) × Am (k) Ui = {([x0 , . . . , xn ], y1 , . . . , ym ) : xi = 0} An (k) × Am (k) and the ideals Ii = I (Ui ∩ S) ⊂ k[w 0 , . . . , w i−1 , w i+1 , . . . , w n , y1 , . . . , ym ]. If Ji denotes the homogenization of Ii relative to w 0 , . . . , w i−1 , w i+1 , . . . , w n (using xi as the homogenizing variable) then n J (S) = ∩i=0 Ji .

The proof proceeds just as in Section 9.4.

10.2

Projective elimination ideals Given a closed subset X ⊂ Pn (k) × Am (k), we would like algorithms for computing the image π2 (X ) ⊂ Am (k). Let J ⊂ k[x0 , . . . , xn ] be homogeneous and m = x0 , . . . , xn . The saturation of J is defined

Definition 10.3

J˜ = {F ∈ k[x0 , . . . , xn ] : m N F ⊂ J for some N 0}. An ideal is saturated if it is equal to its saturation. Note that an ideal is irrelevant if and only if its saturation is k[x0 , . . . , xn ]. Generally, we have X ( J˜) = X (J ) (see Exercise 10.2.)

172


Given an ideal J ⊂ k[x0 , . . . , xn , y1 , . . . , ym ], homogeneous in x0 , . . . , xn , the projective elimination ideal is defined

Definition 10.4

Jˆ = { f ∈ k[y1 , . . . , ym ] : m N f ⊂ J for some N 0},

m = x0 , . . . , xn .

The reason for the fudge-factor m N is that we want any irrelevant ideal to have elimination ideal 1 : The image of the closed subset defined by an irrelevant ideal (i.e., the empty set) is empty! Consider an ideal homogeneous in x0 , . . . , xn ,

Proposition 10.5

J ⊂ k[x0 , . . . , xn , y1 , . . . , ym ], defining a closed subset X (J ) ⊂ Pn (k) × Am (k). Then we have π2 (X (J )) ⊂ V ( Jˆ). Proof

Suppose we have (b1 , . . . , bm ) ∈ π2 (X (J )), corresponding to ([a0 , . . . , an ], b1 , . . . , bm ) ∈ X (J ).

For each F ∈ J we have F(a0 , . . . , an , b1 , . . . , bm ) = 0. For some i, ai = 0, say, a0 = 0. If h(y1 , . . . , ym ) ∈ Jˆ then x0N h ∈ F for some N 0, so a0N h(b1 , . . . , bm ) = 0 and hence h(b1 , . . . , bm ) = 0.

Assume k is algebraically closed. Let J ⊂ k[x0 , . . . , xn , y1 , . . . , ym ] be homogeneous in x0 , . . . , xn , X = X (J ) ⊂ Pn (k) × Am (k), and Jˆ ⊂ k[y1 , . . . , ym ] the projective elimination ideal. Then

Theorem 10.6 (Projective Elimination)

π2 (X (J )) = V ( Jˆ). It suffices to show V ( Jˆ) ⊂ π2 (X ): the reverse implication has already been proven. Suppose we are given c = (c1 , . . . , cm ) ∈ V ( Jˆ), but c ∈ π2 (X ). Write J = F1 , . . . , Fr , with F1 , . . . , Fr homogenous in x0 , . . . , xn , and set

Proof

Fi (x0 , . . . , xn , c) = Fi (x0 , . . . , xn , c1 , . . . , cm ). Since the equations F1 (x0 , . . . , xn , c) = . . . = Fr (x0 , . . . , xn , c) = 0 have no common solutions, Projective Nullstellensatz gives m N ⊂ Fi (x0 , . . . , xn , c) i=1,...,r

10.2 PROJECTIVE ELIMINATION IDEALS

173

for some N 0. In particular, for each monomial x α with |α| = N we can write xα =

r

Fi (x0 , . . . , xn , c)Hi,α (x0 , . . . , xn ),

i=1

where the Hi,α are homogeneous. We may choose monomials x β j and indices i j ∈ {1, . . . , r } for j = 1, . . . , NN+n such that {x β j Fi j (x0 , . . . , xn , c)} j=1,...( N +n) N

forms a basis for homogeneous forms of degree N in x0 , . . . , xn . Consider the corresponding polynomials in x0 , . . . , xn , y1 , . . . , ym : G j (x0 , . . . , xn , y1 , . . . , ym ) := x β j Fi j (x0 , . . . , xn , y1 , . . . , ym ). Express Gj =

x α A j,α (y1 , . . . , ym )

|α|=N

so that A = (A j,α ) is an determinant

N +n N

×

N +n N

matrix of polynomials in y1 , . . . , ym . The

D(y1 , . . . , ym ) = det(A) is hence also a polynomial in y1 , . . . , ym , and D(c1 , . . . , cm ) = 0 by hypothesis. By Cramer’s rule α

D(y1 , . . . , ym )x =

N +n ( N )

B j,α (y1 , . . . , ym )G j (x0 , . . . , xm , y1 , . . . , ym )

j=1

for a suitable matrix B = (B j,α ), with entries polynomials in y1 , . . . , ym . It follows that D(y1 , . . . , ym )x α ∈ F1 , . . . , Fr

and thus D(y1 , . . . , ym ) ∈ Jˆ. This contradicts the assumption that (c1 , . . . , cm ) ∈ V ( Jˆ).

This is strikingly similar to our analysis of resultants, especially Theorems 5.5 and 5.8. We encourage the reader to apply the argument for Theorem 10.6 to J = ad x1d + ad−1 x1d−1 x0 + · · · + a0 x0d , be x1e + be−1 x1e−1 x0 + · · · + b0 x0e ,

Remark 10.7

174


which defines a closed subset X ⊂ P1 (k) × Am+n+2 (k). Here a0 , . . . , ad , b0 , . . . , be play the role of the y-variables in our argument; the elimination ideal is generated by resultant of the two generators. One advantage of this special case is the good control we have over N , which makes possible the elegant determinantal form of the resultant. 10.2.1 Caveats for nonclosed fields

The Projective Elimination Theorem needs the assumption that k is algebraically closed. For general fields, at least two problems arise. First, the ideal J (X (J )) may be much larger than J , and may include equations that are not algebraic consequences of the equations in J . This is the case if X (J ) happens to have few points with coordinates in k, i.e., when there are polynomials vanishing on X (J ) that are not consequences of polynomials in J . See Example 3.17 for concrete examples and further discussion. The second problem arises even when X (J ) does have lots of k-rational points. Consider the real variety V = ([x0 , x1 ], y) : x02 = yx12 ⊂ P1 (R) × A1 (R). The image, π2 (V ) = {y ∈ A1 (R) : y ≥ 0}, is not a Zariski-closed subset of A1 (R).

10.3

Computing the projective elimination ideal Let J ⊂ k[x0 , . . . , xn , y1 , . . . , ym ] be an ideal, homogeneous in x0 , . . . , xn . How do we compute the projective elimination ideal Jˆ? Method I

By definition, Jˆ = (∪ N J : m N ) ∩ k[y1 , . . . , ym ]

for m = x0 , . . . , xn . For any particular N , Algorithm 8.10 computes the quotient J : m N . To compute the union, we use the ascending chain condition and a generalization of Proposition 8.34: Proposition 10.8

Let R be a ring and I, J ⊂ R ideals. Suppose that for some

integer M ≥ 0 I : J M = I : J M+1 , where J 0 = R by convention. Then we have I : J M = I : J M+1 = I : J M+2 = I : J M+3 = . . . .

10.4 IMAGES OF PROJECTIVE VARIETIES ARE CLOSED

175

Method II For each i = 0, . . . , n, let Ii denote the dehomogenization of J with respect to xi . Then n Iˆ := ∩i=0 (Ii ∩ k[y1 , . . . , ym ])

is equal to Jˆ. Proof of equivalence

Suppose f ∈ Jˆ, so that for some N 0 we have x0N f, x1N f, . . . , xnN f ∈ J.

It follows that f is contained in each dehomogenization Ii , and thus in Iˆ. We prove the reverse inclusion. Recall that if F is homogeneous in x0 , . . . , xn , f its dehomogenization with respect to xi , and F the rehomogenization of f , then F = xie F for some e. Hence if J is a homogeneous ideal, Ii its dehomogenization with respect to xi , and Ji the rehomogenization with respect to xi , then Ji = J : xiN , for all N 0. We have Ii ∩ k[y1 , . . . , ym ] = Ji ∩ k[y1 , . . . , ym ] and hence n Iˆ = ∩i=0 (Ji ∩ k[y1 , . . . , ym ]).

On the other hand, ∩i=0 Ji = (J : m N ), N 0, so Iˆ = Jˆ.

10.4

Images of projective varieties are closed Assume k is algebraically closed. Let X ⊂ Pn (k) be a projective variety, Y an abstract variety, φ : X → Y a morphism. Then φ(X ) is closed.

Theorem 10.9

Proof

Let φ ⊂ X × Y ⊂ Pn (k) × Y

denote the graph of φ. We claim that this is closed. Consider the induced morphism (φ, Id) : X × Y → Y × Y, with the second factor the identity. We have φ = {(x, y) ∈ X × Y : ( f, Id)(x, y) = (y, y)} = (φ, Id)−1 ( Y ),

176


where Y is the diagonal. Since Y is closed in Y × Y (see Exercise 9.19), φ is closed as well. Choose an affine open covering {V j } for Y ; realize V j ⊂ Am (k) as a closed subset. The intersection φ ∩ π2−1 (V j ) ⊂ Pn (k) × Am (k) is also closed, and the Projective Elimination Theorem (10.6) implies π2 φ ∩ π2−1 (V j ) is closed, which is the equal to π2 ( φ ) ∩ V j = φ(X ) ∩ V j . Thus the intersection of φ(X ) with each affine open neighborhood is closed in that neighborhood, and φ(X ) is closed in Y . Assume k is algebraically closed. Let V ⊂ An (k) be an affine variety, φ : V → Am (k) a morphism with graph φ . Let

Proposition 10.10

φ ⊂ Pn (k) × Am (k) be the projective closure. Then π2 ( φ ) = φ(V ). The Projective Elimination Theorem (10.6) implies π2 ( φ ) is closed, and it contains the image φ(V ), so it also contains φ(V ). On the other hand, π2−1 (φ(V )) contains φ , and also φ ; applying φ2 gives the desired result.

Proof

Let V ⊂ An (k) be affine and φ : V → Am (k) a morphism. If its image fails to be closed, the extra points in the closure come from points ‘at infinity’, i.e.,

Remark 10.11

φ(V ) − φ(V ) ⊂ π2 ( φ ∩ X (x0 )). Here we are identifying An (k) with U0 = {x0 = 0} ⊂ Pn (k); X (x0 ) = (Pn (k) × Am (k)) \ (U0 × Am (k)) is the hyperplane at infinity.

10.5

Further elimination results For the remainder of this section, we discuss how elimination results for affine varieties can be applied in projective contexts. Let X ⊂ Pn (k) be a projective variety and φ : X Pm (k) a rational map induced by the polynomial map

Proposition 10.12

Pn (k) Pm (k) [x0 , . . . , xn ] → [φ0 , . . . , φm ].

10.6 EXERCISES

177

Writing I = y0 − φ0 , . . . , ym − φm + k[x0 , . . . , xn , y0 , . . . , ym ]J (X ) J = I ∩ k[y0 , . . . , ym ]. then X (J ) is the closure of the image of φ. Of course, if k is algebraically closed and φ is a morphism then φ(X ) is automatically closed. Let C(X ) ⊂ An+1 (k) be the cone over X , i.e., the subvariety defined by the equations in J (X ), or the union of lines

Proof

{λ(x0 , . . . , xn ) : λ ∈ k} ⊂ An+1 (k) for [x0 , . . . , xn ] ∈ X . The polynomials defining φ also determine a map of affine varieties ϕ : C(X ) → Am+1 (k), whose graph is defined by the equations in I . It is easy to see that C(φ(X )) = ϕ(C(X )), which has equations that may be analyzed with affine elimination theory (see §4.1). In particular, the image is given by the ideal J .

10.6

Exercises

10.1 10.2

Prove Proposition 10.8. Let J ⊂ k[x0 , . . . , xn ] be a homogeneous ideal with saturation J˜. (a) Show that X (J ) = X ( J˜). (b) For each i = 0, . . . , n, let Ii ⊂ k[y0 , . . . , yi−1 , yi+1 , . . . , yn ] denote the dehomogenization of J and Ji ⊂ k[x0 , . . . , xn ] the homogenization of Ii . Show that J˜ = J0 ∩ J1 ∩ . . . ∩ Jn .

10.3

(c) Show that J = J˜ if and only if m = x0 , . . . , xn is not an associated prime of J . Consider the ideal J = x0 y0 + x1 y1 , x0 y1 + x1 y0 ⊂ C[x0 , x1 , y0 , y1 ], which is homogeneous in x0 , x1 . (a) Compute the intersection J ∩ C[y0 , y1 ]. (b) Compute the projective elimination ideal Jˆ.

178


10.4

(c) Compute the image π2 (X (J )), where X (J ) ⊂ P1 (C) × A2 (C). (d) Note that J is also homogeneous in y0 and y1 and thus determines a closed subset X ⊂ P1 (C) × P1 (C). Describe X . (Cayley cubic surface) Consider the rational map ρ : P2 (k) P3 (k) taking [x0 , x1 , x2 ] to [−x0 x1 x2 , x0 x1 (x0 + x1 + x2 ), x0 x2 (x0 + x1 + x2 ), x1 x2 (x0 + x1 + x2 )].

10.5

Describe the indeterminacy of ρ in P2 (k) and the equations of the image in P3 (k). Optional: Show that P2 (k) image(ρ) is birational. F ∈ k[x0 , . . . , xn , y0 , . . . , ym ] is bihomogeneous of bidegree (d, e) if it is homogeneous in x0 , . . . , xn of degree d and homogeneous in y0 , . . . , ym of degree e. (a) Show that if F is bihomogeneous then Y (F) = {([a0 , . . . , an ], [b0 , . . . , bm ]) : F(a0 , . . . , an , b0 , . . . , bm ) = 0} is a well-defined closed subset of Pn (k) × Pm (k). (b) Using the Segre embedding, express Y (F) as the locus in Pmn+m+n (k) where a collection of homogeneous forms vanish. Hint: If d ≥ e express y β F, for each β with |β| = d − e, as a polynomial in the products xi y j . (c) For F = x02 y0 + x12 y1 + x22 y2 write down explicit equations for Y (F) ⊂ P8 (k). (d) If J ⊂ k[x0 , . . . , xn , y0 , . . . , ym ] is a bihomogeneous ideal show that Y (J ) := ∩ F∈J Y (F) ⊂ Pn (k) × Pm (k)

10.6

is projective. (e) Let X ⊂ Pm (k) and Y ⊂ Pn (k) be projective varieties. Show that X × Y is projective. Given a polynomial morphism φ : Pn (k) → Pm (k) [x0 , . . . , xn ] → [φ0 , . . . , φm ] it can be tricky to extract equations for the graph. (a) Show that the graph always contains the bihomogeneous equations yi φ j − y j φi . (b) For the Veronese morphism ν(2) : P1 (k) → P2 (k)

[x0 , x1 ] → x02 , x0 x1 , x12

10.6 EXERCISES

179

show that the equations of the graph are y0 x1 − y1 x0 , y1 x1 − y2 x0 .

10.7

(c) Extract bihomogenous equations for the graph of ν(2) : P2 (k) → P5 (k). A binary form of degree d is a nonzero homogeneous polynomial F = a0 x0d + a1 x0d−1 x1 + · · · + ad x1d ∈ k[x0 , x1 ]. Binary forms up to scalar multiples are parametrized by the projective space P(k[x0 , x1 ]d ) = {[a0 , . . . , ad ]}. (a) Show that multiplication induces a morphism P(k[x0 , x1 ]e ) × P(k[x0 , x1 ]d−e ) → P(k[x0 , x1 ]d ) (F, G) → F G. Suppose k is algebraically closed. We say F has a root of multiplicity ≥ e if there exists a nonzero linear form L = l0 x0 + l1 x1 with L e |F. (b) Prove that the binary forms with a root of multiplicity e form a closed subset Re ⊂ P(k[x0 , x1 ]d ). Hint: Verify that the map P(k[x0 , x1 ]1 ) × P(k[x0 , x1 ]d−e ) → P(k[x0 , x1 ]d ) (L , G) → L e G

10.8

is a morphism of projective varieties. (c) Write down explicit equations for r R1 ⊂ Pd (k); r R2 ⊂ P3 (k), P4 (k). Let F be homogeneous of degree d in k[x0 , . . . , xn ]; all such forms (up to scalars) are parametrized by n+d P( d )−1 (k) = P(k[x0 , . . . , xn ]d ).

For a = [a0 , . . . , an ] ∈ Pn (k), we say that F has multiplicity ≥ e at a (or vanishes to order e at a) if (cf. Exercise 9.2): F ∈ J (a)e = xi a j − x j ai ,

i, j = 0, . . . , n e .

(a) Show that F has multiplicity ≥ e at [1, 0, 0, . . . , 0] if and only if the dehomogenization μ0 (F) ∈ k[y1 , . . . , yn ] has no terms of degree < e. Consider the locus n+d Z e := {(a, F) : F has multiplicity ≥ e at a } ⊂ Pn (k) × P( d )−1 (k).

Assume k is a field of characteristic other than two and d = 2.

180


(b) Show that Z 2 = {(a, F) : ∂ F/∂ xi (a0 , . . . , an ) = 0, i = 0, . . . , n}. (c) Write y00 ⎜ y01 ⎜ F(x0 , . . . , xn ) = (x0 x1 · · · xn ) ⎜ . ⎝ ..

y01 y11 .. .

... ... .. .

⎞⎛ ⎞ x0 y0n ⎜ x1 ⎟ y1n ⎟ ⎟⎜ ⎟ .. ⎟ ⎜ .. ⎟ . ⎠⎝ . ⎠

y0n

y1n

···

ynn

⎛

xn

for Y = (yi j ) a symmetric (n + 1) × (n + 1) matrix. Compute the projective elimination ideal of ∂ F/∂ x0 , . . . , ∂ F/∂ xn ⊂ k[x0 , . . . , xn , y00 , . . . , ynn ] and show that π2 (Z 2 ) = {Y : det(Y ) = 0}. (d) Returning to the general case, show that Z e is closed. Conclude that the hypersurfaces with point of multiplicity ≥ e are closed in n+d P( d )−1 (k).

11 Parametrizing linear subspaces

The vector subspaces of k N are parametrized by a projective variety called the Grassmannian. This is the first instance of a very important principle: algebraic varieties with common properties are often themselves classified by an algebraic variety. Applying the techniques of algebraic geometry to this ‘classifying variety’ gives rise to rich insights. For example, we write down explicit equations for the Grassmannian in projective space, using the formalism of exterior algebra. Such representations are crucial for many applications.

11.1

Dual projective spaces Recall that we defined projective space P n (k) = space of all lines 0 ∈ ⊂ k n+1 = one-dimensional subspaces = span(x0 , . . . , xn ). The dual projective space Pˇ n (k) is the space of all n-dimensional vector subspaces H ⊂ k n+1 .

Definition 11.1

It is a basic fact of linear algebra that every n-dimensional subspace can be expressed as H = H ( p0 , . . . , pn ) = {(x0 , . . . , xn ) : p0 x0 + · · · + pn xn = 0} for some ( p0 , . . . , pn ) = 0, where H ( p0 , . . . , pn ) = H ( p0 , . . . , pn ) ⇔ [ p0 , . . . , pn ] = [ p0 , . . . , pn ] ∈ P n (k). Thus the map H ( p0 , . . . , pn ) → [ p0 , . . . , pn ] allows us to identify P n (k) with Pˇ n (k).

181

182

PARAMETRIZING LINEAR SUBSPACES

Example 11.2 (General incidence correspondence)

Consider the locus

W := {(, H ) : ⊂ H ⊂ k n+1 } = {[x0 , . . . , xn ], [ p0 , . . . , pn ] : x0 p0 + · · · + xn pn = 0} ⊂ P n (k) × Pˇ n (k). Recall the Segre embedding φ : P n (k) × Pˇ n (k) → Pn +2n (k) k[z 00 , . . . , z nn ] → k[x0 , . . . , xn ] × k[ p0 , . . . , pn ] z i j → xi p j 2

with image given by the vanishing of the 2 × 2 minors of the matrix ⎛

z 00 ⎜ z 10 ⎜ Z =⎜ . ⎝ ..

z 01 z 11 .. .

z n0

z n1

... ... ... ...

⎞ z 0n z 1n ⎟ ⎟ ⎟. ...⎠ z nn

The locus W ⊂ φ(P n (k) × Pˇ n (k)) is defined by z 00 + · · · + z nn = Trace(Z ) = 0, and thus is a projective variety.

11.2

Tangent spaces and dual varieties Let V ⊂ An (k) be a hypersurface, i.e., I (V ) = g for some g ∈ k[y1 , . . . , yn ]. Given b ∈ V , the affine tangent space is defined: Tb V = (y1 , . . . , yn ) :

n

∂g/∂ y j |b · (y j − b j ) = 0 .

(11.1)

j=1

This is an affine-linear subspace of An (k). It is a hyperplane if ∂g/∂ y j |b = 0 for some index j; in this case, we say that V is smooth at b. Otherwise, it is singular. For a general affine V ⊂ An (k) and b ∈ V , we define the affine tangent space by Tb V = (y1 , . . . , yn ) :

n

∂g/∂ y j |b (y j − b j ) = 0 for each g ∈ I (V ) .

j=1

We say that V is smooth at b if it has a unique irreducible component containing b of dimension dim Tb V . (The dimension of a variety at a point is well-defined when the variety has a unique irreducible component containing the point.)

11.2 TANGENT SPACES AND DUAL VARIETIES

183

Consider an affine linear subspace expressed as the solutions to a system of linear equations, i.e., ⎫ ⎧ ⎛ ⎞ y1 ⎪ ⎪ ⎬ ⎨ ⎜ .. ⎟ L = y = ⎝ . ⎠ : Ay = b ⊂ An (k) ⎪ ⎪ ⎭ ⎩ yn where A is an m × n matrix of maximal rank, b ∈ k m is a column vector, and dim L = n − m. Express An (k) as a distinguished open subset U0 ⊂ P n (k). Using Proposition 9.16, the projective closure of L can be expressed ⎫ ⎧ ⎛ ⎞ x1 ⎪ ⎪ ⎬ ⎨ ⎜ .. ⎟ n L = [x0 , . . . , xn ] ∈ P (k) : A ⎝ . ⎠ = bx0 . ⎪ ⎪ ⎭ ⎩ xn This is a linear subspace of P n (k) as well. Let X ⊂ P n (k) be a hypersurface with J (X ) = F , where F ∈ k[x0 , . . . , xn ] is homogeneous of degree d, and a = [a0 , . . . , an ] ∈ X ∩ Ui . Write V = X ∩ Ui , b = (a0 /ai , . . . , ai−1 /ai , ai+1 /ai , . . . , an /ai ) ∈ An (k) the corresponding point of affine space, and f ∈ k[y0 , . . . , yi−1 , yi+1 , . . . , ym ] the dehomogenization of F with respect to xi . Then the projective closure Tb V equals the linear subspace

n Ta X := [x0 , . . . , xn ] : (∂ F/∂ xi )|a xi = 0 .

Proposition 11.3

i=0

This is called the projective tangent space of X at a. In particular, a hypersurface in P n (k) is singular at a point if its projective tangent space there is P n (k). Proof

For notational simplicity assume i = 0. Let μ0 : k[x0 , . . . , xn ] → k[y1 , . . . , ym ]

denote the dehomogenization homomorphism. For i = 1, . . . , n we have μ0 (∂ F/∂ xi ) = ∂ f /∂ yi . Writing out f as a sum of homogeneous pieces f = f0 + f1 + · · · + fd , we find μ0 (∂ F/∂ x0 ) = d f 0 + (d − 1) f 1 + · · · + f d−1 .

184


We analyze the expression n

(∂ F/∂ xi )|a xi .

(11.2)

i=0

This is homogeneous in a of degree d − 1, i.e., n

(∂ F/∂ xi )|λa = λd−1

i=0

n

(∂ F/∂ xi )|a .

i=0

Since we are only interested in where this vanishes, we may assume that a0 = 1 and ai = bi , i = 1, . . . , n. Dehomogenizing (11.2) therefore yields (d f 0 + (d − 1) f 1 + · · · + f d−1 )|b +

n

∂ f /∂ y j |b y j .

j=1

This is equal to (11.1) provided we can establish −

n

b j ∂ f /∂ y j |b = (d f 0 + (d − 1) f 1 + · · · + f d−1 )|b .

j=1

If F ∈ k[x0 , . . . , xn ] is homogeneous then

Lemma 11.4 (Euler’s Formula)

deg(F) · F =

n

xi ∂ F/∂ xi .

i=0

Both sides are linear in F. It suffices then to check the formula for x α = x0α0 . . . xnαn , α0 + · · · + αn = d. In this case we have

Proof of lemma

n i=0

xi

n ∂ α x = xα αi = d x α . ∂ xi i=0

Applying μ0 to Euler’s formula and evaluating at (b1 , . . . , bn ) yields d · f (b1 , . . . , f n ) = (d f 0 + (d − 1) f 1 + · · · + f d−1 )|b +

n

b j ∂ f /∂ y j |b .

j=1

Since f (b1 , . . . , bn ) = 0 we obtain (d f 0 + (d − 1) f 1 + · · · + f d−1 )|b = −

n

b j ∂ f /∂ y j |b .

j=1

Corollary 11.5

1. 2.

Retain the notation of Proposition 11.3.

If a ∈ X ∩ Ui and ∂ F/∂ x j |a = 0 for each j = i then ∂ F/∂ xi |a = 0 as well. a ∈ X is singular if and only if ∂ F/∂ xi |a = 0 for each i = 0, . . . , n.

11.2 TANGENT SPACES AND DUAL VARIETIES

3.

185

Suppose that (char(k), d) = 1 or char(k) = 0. Singular points a ∈ X are precisely the simultaneous solutions to the equations ∂ F/∂ x0 = . . . = ∂ F/∂ xn = 0. Proof

Only the last assertion requires proof. Euler’s formula gives F(a0 , . . . , an ) = 1/d

n

ai ∂ F/∂ xi |(a0 ,...,an ) ,

i=0

which vanishes whenever the partials all vanish.

Let X ⊂ P n (k) be a projective variety and a ∈ X . The projective tangent space of X at a is defined

n Ta X = [x0 , . . . , xn ] : ∂ F/∂ xi |a xi = 0 for each F ∈ J (X ) ⊂ P n (k).

Definition 11.6

i=0

X is smooth at a if it has a unique irreducible component containing a of dimension dim Ta X . Let X ⊂ P n (k) be an irreducible projective variety. The dual variety Xˇ ⊂ Pˇ (k) is the closure of the locus of all hyperplanes tangent to X at smooth points, i.e.,

Definition 11.7 n

Xˇ = {H ∈ Pˇ n (k) : Ta X ⊂ H for some a ∈ X smooth }. Let k be algebraically closed and X ⊂ P n (k) projective. Consider the incidence variety

Remark 11.8

W X = {(a, H ) : a ⊂ H ⊂ Ta X } ⊂ X × Pˇ n (k) ⊂ P n (k) × Pˇ n (k), which is contained in the incidence correspondence introduced in Example 11.2. Note that π2 (W X ) is closed by Theorem 10.9; Xˇ is a union of irreducible components of this variety. For many applications, it is important to restrict attention to hypersurfaces without singularities. However, a hypersurface with no singularities over a given field may acquire them after the field is extended. This is just the problem we faced in defining a morphism. The following definition circumvents this difficulty: A hypersurface X (F) ⊂ P n (k) is smooth if F, ∂ F/∂ x0 , . . . , ∂ F/∂ xn is irrelevant.

Definition 11.9

186


Proposition 11.10

Let X ⊂ P n (k) be a hypersurface with J (X ) = F . The poly-

nomial map P n (k) Pˇ n (k) [x0 , . . . , xn ] → [∂ F/∂ x0 , . . . , ∂ F/∂ xn ] induces a morphism X → P n (k) when X is smooth. Even when X is singular its image is Xˇ , with equations given by F, p0 − ∂ F/∂ x0 , . . . , pn − ∂ F/∂ xn ∩ k[ p0 , . . . , pn ]. The first statement follows from general properties of polynomial maps of projective spaces (Proposition 9.30). Let X s ⊂ X denote the union of the irreducible components containing smooth points. We obtain a rational map X s P n with closed image Xˇ . The equations are obtained from Proposition 10.12.

Proof

Example 11.11

Assume char(k) = 3. Consider the smooth plane curve X = (x0 , x1 , x2 ) : x03 + x13 + x23 = 0 ⊂ P2 (k).

Then the dual is given as Xˇ = ( p0 , p1 , p2 : p06 + p16 + p26 − 2 p03 p23 − 2 p13 p23 − 2 p13 p03 ⊂ Pˇ 2 (k). 11.2.1 Plücker formulas

When X is a smooth hypersurface the dual Xˇ is usually singular, even for plane curves! Figure 11.1 shows two typical cases where the dual curve acquires singularities; for generic X these are only possibilities. In the first case we have an inflectional tangent, where the tangent line meets X at p with multiplicity 3 rather than 2. (Formal definitions can be found in §12.4.2.) X admits an inflectional tangent at p precisely when the differential of the map X → Pˇ 2 vanishes at p. One can show that the image Xˇ has a cusp at , i.e., a singularity with local normal form y 2 = x 3 . In the second case, we have a bitangent, i.e. is tangent to X in two points p1 , p2 . Thus p1 and p2 are mapped to the same point of Pˇ 2 , so the image Xˇ has two local branches at these points. We say that Xˇ has a node at , with local normal form y 2 = x 2 . There are formulas relating the invariants of X and Xˇ . Let d and dˇ be the degrees of X and Xˇ , f the number of inflectional tangents X (which equals the number of cusps of Xˇ ), and b the number of bitangents to X (the number of nodes of Xˇ ). Then we have the Plücker formulas dˇ = d(d − 1) f = 3d(d − 2) ˇ dˇ − 1) − d − 3 f )/2. b = (d( We will deduce the second formula from the Bezout Theorem in §12.4.2.

11.3 GRASSMANNIANS: ABSTRACT APPROACH

187

dual to X

inflectional tangent

cusp singularity X

bitangent dual to X X node singularity Figure 11.1

11.3

Typical singularities of the dual curve.

Grassmannians: Abstract approach We have defined the spaces of all lines and hyperplanes in k n+1 . Why not consider subspaces of arbitrary dimension? The study of arbitrary finite-dimensional spaces was pioneered by Herman Günter Grassmann (1809–1877) in his 1844 book Die Lineale Ausdehnungslehre, ein neuer Zweig der Mathematik. (An English translation is available in [13].) For each M = 1, 2, . . . , N − 1, the Grassmannian Gr(M, N ) is the set of all vector subspaces ⊂ k N of dimension M.

Definition 11.12

In particular, Gr(1, N ) = P N −1 (k) and Gr(N − 1, N ) = Pˇ N −1 (k). The Grassmannian Gr(M, N ) carries the structure of an abstract variety. It is irreducible and rational of dimension M(N − M).

Theorem 11.13

For the moment, we only describe the affine open covering and the gluing maps. Here is the idea of the argument: fix a basis e1 , . . . , e N for k N . Each can be expressed as the row-space of an M × N matrix W of maximal rank M. This W is not at all unique. Indeed, applying elementary row operations to W does not affect

188


, so we may replace W by its reduced row echelon form. For ‘most’ W , we have ⎛

1

⎜ RREF(W ) = ⎜ ⎝0 .. .

0 .. .

...

0

...

⎞

0

b1M+1 .. .

b1N

...

1

b M M+1

. . . bM N

⎟ ... ⎟ ⎠ = ( I B ),

where I is the identity. Allowing permutations of e1 , . . . , e N , every subspace admits a reduced row echelon form of this type. To formalize this, we will need several results from linear algebra: Lemma 11.14

Fix a partition

{1, . . . , N } = S ∪ T,

S = {s1 , . . . , s M }, T = {t1 , . . . , t N −M }

with s1 < . . . < s M , t1 < . . . < t N −M . For each N × (M − N ) matrix B = (bst ) with rows and columns indexed by S and T respectively, consider the subspace (S; B) = span es + bst et : s ∈ S . t∈T

These satisfy the following:

r dim( (S; B)) = M; r (S; B) = (S; C) only if B = C. Let R(S; B) be the M × N matrix with ith row equal to esi + For example, when S = {1, . . . , M} we have

Proof

⎛

1

⎜ R(S; B) = ⎝ 0

0 .. .

... .. .

b1M+1 .. .

0

...

1

b M M+1

...

b1N

t∈T

bsi t et .

⎞

⎟ . . . . . . ⎠ = ( I B ), . . . bM N

where I is an M × M identity matrix. The rows of R(S; B) span (S; B); since R(S; B) has rank M, dim( (S; B)) = M and the first assertion follows. We leave it to the reader to verify the second assertion. Lemma 11.15

Retain the notation of the previous lemma. Each of the distinguished

subsets U S = { (S; B) : B is a M × (N − M) matrix } ⊂ Gr(M, N ) admits a natural identification ∼

ψ S : U S → A M(N −M) (k) (S; B) → B.

11.3 GRASSMANNIANS: ABSTRACT APPROACH

189

The Grassmannian is covered by these subsets, i.e., Gr(M, N ) = ∪partitions S∪T U S . Each ∈ Gr(M, N ) can be expressed as span(w 1 , . . . , w M ) for suitable linearly independent w 1 , . . . , w M ∈ ⊂ k N . Let W be the M × N matrix having these vectors as its rows. Note that multiplication from the left by an invertible matrix does not change the span of the rows, i.e.,

Proof

row span(W ) = row span(AW ) provided det(A) = 0. Choose indices s1 , . . . , s M such that the corresponding columns of W are linearly independent and let W S be the square matrix built from these columns. The matrix W S−1 W is of the form R(S; B) for some M × (N − M) matrix B. The natural identification with affine space arises by identifying M × (N − M) matrices with points in A M(N −M) (k). Lemma 11.16

Given partitions corresponding to S, S ⊂ {1, . . . , N }, the overlap

maps ρ S S := ψ S ◦ ψ S−1 : A M(N −M) (k) A M(N −M) (k) are birational, given by the rule ψ S−1

ψS

B → (S; B) = (S ; B ) → B R(S; B) → R(S; B)−1 S R(S; B). Here R(S; B) S is the M × M matrix obtained by extracting the columns of R(S; B) indexed by S ; this is invertible provided B ∈ ψ S (U S ∩ U S ). Observe that R(S; B)−1 S R(S; B) contains the identity matrix in the columns indexed by S . It is therefore of the form R(S ; B) for some suitable B . Again we have

Proof

row span(R(S; B)) = row span(R(S ; B )), i.e., (S; B) and (S ; B ) coincide.

What is left to do in the proof of Theorem 11.13? We have not verified that the ρ S S satisfy the closed-graph condition. In Proposition 11.30, we will establish that the Grassmannian is projective, in a way that is compatible with the proposed abstract variety structure: our distinguished open subsets U S will be intersections of the Grassmannian with distinguished open subsets of the ambient projective space.

190


We give examples of gluing maps for Gr(2, 4). Let S = {1, 2} and

Example 11.17

S = {1, 3} so that U S = span(e1 + b13 e3 + b14 e4 , e2 + b23 e3 + b24 e4 ) A4 (k), U S = span(e1 + b12 e2 + b14 e4 , e3 + b32 e2 + b34 e4 ) A4 (k). Start with the matrix R(S; B) =

1 0 b13 0 1 b23

b14 b24

and left multiply by R(S; B)−1 S =

1 −b13 /b23 0 1/b23

to get −1 R(S ; B ) = R S; B S =

1 −b13 /b23 0 1/b23

0 1

b14 − b13 b24 /b23 b24 /b23

.

The gluing map is ρ S∗ S b12 = −b13 /b23 , ∗ ρ S S b32 = 1/b23 ,

ρ S∗ S b14 = b14 − b13 b24 /b23 ∗ ρ S S b34 = b24 /b23 .

We have seen that P N −1 (k) Pˇ N −1 (k); this is not a coincidence: Proposition 11.18

Choose a nondegenerate inner product on k N , e.g., the standard

dot-product (a1 , . . . , a N ) · (b1 , . . . , b N ) = a1 b1 + · · · + a N b N . These we have a natural identification Gr(M, N ) Gr(N − M, N ). Proof

Given a subspace , we define the orthogonal complement ⊥ = {x ∈ k N : x · v = 0 for each v ∈ }.

Since the product is nondegenerate, dim ⊥ = N − dim and ( ⊥ )⊥ = . The association Gr(M, N ) → Gr(N − M, N ) → ⊥ gives the desired identification.

11.4 EXTERIOR ALGEBRA

11.4

191

Exterior algebra In the last section, we introduced the Grassmannian as an abstract variety. Both for theoretical and practical reasons, it is very useful to have a concrete realization of the Grassmannian in projective space. Here we develop the algebraic formalism needed to write down its homogeneous equations. We work over a field with characteristic char(k) = 2.

11.4.1 Basic definitions

Let V = {c1 e1 + · · · + c N e N } be a finite-dimensional vector space with basis {e1 , . . . , e N }. For each M = 0, . . . , N , the Mth exterior power is defined as the vector space

M V = ci1 ...i M ei1 ∧ ei2 . . . ∧ ei M , 1≤i 1

Introduction to Algebraic Geometry

Introduction to algebraic geometry

Introduction to algebraic geometry